arch/x86/kernel/traps.c

   1 /*
   2  *  Copyright (C) 1991, 1992  Linus Torvalds
   3  *  Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
   4  *
   5  *  Pentium III FXSR, SSE support
   6  *      Gareth Hughes <gareth@valinux.com>, May 2000
   7  */
   8
   9 /*
  10  * Handle hardware traps and faults.
  11  */
  12 #include <linux/interrupt.h>
  13 #include <linux/kallsyms.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/kprobes.h>
  16 #include <linux/uaccess.h>
  17 #include <linux/utsname.h>
  18 #include <linux/kdebug.h>
  19 #include <linux/kernel.h>
  20 #include <linux/module.h>
  21 #include <linux/ptrace.h>
  22 #include <linux/string.h>
  23 #include <linux/delay.h>
  24 #include <linux/errno.h>
  25 #include <linux/kexec.h>
  26 #include <linux/sched.h>
  27 #include <linux/timer.h>
  28 #include <linux/init.h>
  29 #include <linux/bug.h>
  30 #include <linux/nmi.h>
  31 #include <linux/mm.h>
  32 #include <linux/smp.h>
  33 #include <linux/io.h>
  34
  35 #ifdef CONFIG_EISA
  36 #include <linux/ioport.h>
  37 #include <linux/eisa.h>
  38 #endif
  39
  40 #ifdef CONFIG_MCA
  41 #include <linux/mca.h>
  42 #endif
  43
  44 #if defined(CONFIG_EDAC)
  45 #include <linux/edac.h>
  46 #endif
  47
  48 #include <asm/stacktrace.h>
  49 #include <asm/processor.h>
  50 #include <asm/debugreg.h>
  51 #include <asm/atomic.h>
  52 #include <asm/system.h>
  53 #include <asm/traps.h>
  54 #include <asm/desc.h>
  55 #include <asm/i387.h>
  56
  57 #include <asm/mach_traps.h>
  58
  59 #ifdef CONFIG_X86_64
  60 #include <asm/pgalloc.h>
  61 #include <asm/proto.h>
  62 #else
  63 #include <asm/processor-flags.h>
  64 #include <asm/setup.h>
  65 #include <asm/traps.h>
  66
  67 #include "cpu/mcheck/mce.h"
  68
  69 asmlinkage int system_call(void);
  70
  71 /* Do we ignore FPU interrupts ? */
  72 char ignore_fpu_irq;
  73
  74 /*
  75  * The IDT has to be page-aligned to simplify the Pentium
  76  * F0 0F bug workaround.. We have a special link segment
  77  * for this.
  78  */
  79 gate_desc idt_table[256]
  80         __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
  81 #endif
  82
  83 DECLARE_BITMAP(used_vectors, NR_VECTORS);
  84 EXPORT_SYMBOL_GPL(used_vectors);
  85
  86 static int ignore_nmis;
  87
  88 static inline void conditional_sti(struct pt_regs *regs)
  89 {
  90         if (regs->flags & X86_EFLAGS_IF)
  91                 local_irq_enable();
  92 }
  93
  94 static inline void preempt_conditional_sti(struct pt_regs *regs)
  95 {
  96         inc_preempt_count();
  97         if (regs->flags & X86_EFLAGS_IF)
  98                 local_irq_enable();
  99 }
 100
 101 static inline void conditional_cli(struct pt_regs *regs)
 102 {
 103         if (regs->flags & X86_EFLAGS_IF)
 104                 local_irq_disable();
 105 }
 106
 107 static inline void preempt_conditional_cli(struct pt_regs *regs)
 108 {
 109         if (regs->flags & X86_EFLAGS_IF)
 110                 local_irq_disable();
 111         dec_preempt_count();
 112 }
 113
 114 #ifdef CONFIG_X86_32
 115 static inline void
 116 die_if_kernel(const char *str, struct pt_regs *regs, long err)
 117 {
 118         if (!user_mode_vm(regs))
 119                 die(str, regs, err);
 120 }
 121 #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         /* Re-enable the IOCK line, wait for a few seconds */
 350         reason = (reason & 0xf) | 8;
 351         outb(reason, 0x61);
 352
 353         i = 2000;
 354         while (--i)
 355                 udelay(1000);
 356
 357         reason &= ~8;
 358         outb(reason, 0x61);
 359 }
 360
 361 static notrace __kprobes void
 362 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
 363 {
 364         if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
 365                         NOTIFY_STOP)
 366                 return;
 367 #ifdef CONFIG_MCA
 368         /*
 369          * Might actually be able to figure out what the guilty party
 370          * is:
 371          */
 372         if (MCA_bus) {
 373                 mca_handle_nmi();
 374                 return;
 375         }
 376 #endif
 377         printk(KERN_EMERG
 378                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 379                         reason, smp_processor_id());
 380
 381         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 382         if (panic_on_unrecovered_nmi)
 383                 panic("NMI: Not continuing");
 384
 385         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 386 }
 387
 388 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
 389 {
 390         unsigned char reason = 0;
 391         int cpu;
 392
 393         cpu = smp_processor_id();
 394
 395         /* Only the BSP gets external NMIs from the system. */
 396         if (!cpu)
 397                 reason = get_nmi_reason();
 398
 399         if (!(reason & 0xc0)) {
 400                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 401                                                                 == NOTIFY_STOP)
 402                         return;
 403 #ifdef CONFIG_X86_LOCAL_APIC
 404                 /*
 405                  * Ok, so this is none of the documented NMI sources,
 406                  * so it must be the NMI watchdog.
 407                  */
 408                 if (nmi_watchdog_tick(regs, reason))
 409                         return;
 410                 if (!do_nmi_callback(regs, cpu))
 411                         unknown_nmi_error(reason, regs);
 412 #else
 413                 unknown_nmi_error(reason, regs);
 414 #endif
 415
 416                 return;
 417         }
 418         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 419                 return;
 420
 421         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 422         if (reason & 0x80)
 423                 mem_parity_error(reason, regs);
 424         if (reason & 0x40)
 425                 io_check_error(reason, regs);
 426 #ifdef CONFIG_X86_32
 427         /*
 428          * Reassert NMI in case it became active meanwhile
 429          * as it's edge-triggered:
 430          */
 431         reassert_nmi();
 432 #endif
 433 }
 434
 435 dotraplinkage notrace __kprobes void
 436 do_nmi(struct pt_regs *regs, long error_code)
 437 {
 438         nmi_enter();
 439
 440         inc_irq_stat(__nmi_count);
 441
 442         if (!ignore_nmis)
 443                 default_do_nmi(regs);
 444
 445         nmi_exit();
 446 }
 447
 448 void stop_nmi(void)
 449 {
 450         acpi_nmi_disable();
 451         ignore_nmis++;
 452 }
 453
 454 void restart_nmi(void)
 455 {
 456         ignore_nmis--;
 457         acpi_nmi_enable();
 458 }
 459
 460 /* May run on IST stack. */
 461 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
 462 {
 463 #ifdef CONFIG_KPROBES
 464         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 465                         == NOTIFY_STOP)
 466                 return;
 467 #else
 468         if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
 469                         == NOTIFY_STOP)
 470                 return;
 471 #endif
 472
 473         preempt_conditional_sti(regs);
 474         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 475         preempt_conditional_cli(regs);
 476 }
 477
 478 #ifdef CONFIG_X86_64
 479 /*
 480  * Help handler running on IST stack to switch back to user stack
 481  * for scheduling or signal handling. The actual stack switch is done in
 482  * entry.S
 483  */
 484 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 485 {
 486         struct pt_regs *regs = eregs;
 487         /* Did already sync */
 488         if (eregs == (struct pt_regs *)eregs->sp)
 489                 ;
 490         /* Exception from user space */
 491         else if (user_mode(eregs))
 492                 regs = task_pt_regs(current);
 493         /*
 494          * Exception from kernel and interrupts are enabled. Move to
 495          * kernel process stack.
 496          */
 497         else if (eregs->flags & X86_EFLAGS_IF)
 498                 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
 499         if (eregs != regs)
 500                 *regs = *eregs;
 501         return regs;
 502 }
 503 #endif
 504
 505 /*
 506  * Our handling of the processor debug registers is non-trivial.
 507  * We do not clear them on entry and exit from the kernel. Therefore
 508  * it is possible to get a watchpoint trap here from inside the kernel.
 509  * However, the code in ./ptrace.c has ensured that the user can
 510  * only set watchpoints on userspace addresses. Therefore the in-kernel
 511  * watchpoint trap can only occur in code which is reading/writing
 512  * from user space. Such code must not hold kernel locks (since it
 513  * can equally take a page fault), therefore it is safe to call
 514  * force_sig_info even though that claims and releases locks.
 515  *
 516  * Code in ./signal.c ensures that the debug control register
 517  * is restored before we deliver any signal, and therefore that
 518  * user code runs with the correct debug control register even though
 519  * we clear it here.
 520  *
 521  * Being careful here means that we don't have to be as careful in a
 522  * lot of more complicated places (task switching can be a bit lazy
 523  * about restoring all the debug state, and ptrace doesn't have to
 524  * find every occurrence of the TF bit that could be saved away even
 525  * by user code)
 526  *
 527  * May run on IST stack.
 528  */
 529 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
 530 {
 531         struct task_struct *tsk = current;
 532         unsigned long condition;
 533         int si_code;
 534
 535         get_debugreg(condition, 6);
 536
 537         /*
 538          * The processor cleared BTF, so don't mark that we need it set.
 539          */
 540         clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
 541         tsk->thread.debugctlmsr = 0;
 542
 543         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 544                                                 SIGTRAP) == NOTIFY_STOP)
 545                 return;
 546
 547         /* It's safe to allow irq's after DR6 has been saved */
 548         preempt_conditional_sti(regs);
 549
 550         /* Mask out spurious debug traps due to lazy DR7 setting */
 551         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 552                 if (!tsk->thread.debugreg7)
 553                         goto clear_dr7;
 554         }
 555
 556 #ifdef CONFIG_X86_32
 557         if (regs->flags & X86_VM_MASK)
 558                 goto debug_vm86;
 559 #endif
 560
 561         /* Save debug status register where ptrace can see it */
 562         tsk->thread.debugreg6 = condition;
 563
 564         /*
 565          * Single-stepping through TF: make sure we ignore any events in
 566          * kernel space (but re-enable TF when returning to user mode).
 567          */
 568         if (condition & DR_STEP) {
 569                 if (!user_mode(regs))
 570                         goto clear_TF_reenable;
 571         }
 572
 573         si_code = get_si_code(condition);
 574         /* Ok, finally something we can handle */
 575         send_sigtrap(tsk, regs, error_code, si_code);
 576
 577         /*
 578          * Disable additional traps. They'll be re-enabled when
 579          * the signal is delivered.
 580          */
 581 clear_dr7:
 582         set_debugreg(0, 7);
 583         preempt_conditional_cli(regs);
 584         return;
 585
 586 #ifdef CONFIG_X86_32
 587 debug_vm86:
 588         /* reenable preemption: handle_vm86_trap() might sleep */
 589         dec_preempt_count();
 590         handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
 591         conditional_cli(regs);
 592         return;
 593 #endif
 594
 595 clear_TF_reenable:
 596         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 597         regs->flags &= ~X86_EFLAGS_TF;
 598         preempt_conditional_cli(regs);
 599         return;
 600 }
 601
 602 #ifdef CONFIG_X86_64
 603 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 604 {
 605         if (fixup_exception(regs))
 606                 return 1;
 607
 608         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 609         /* Illegal floating point operation in the kernel */
 610         current->thread.trap_no = trapnr;
 611         die(str, regs, 0);
 612         return 0;
 613 }
 614 #endif
 615
 616 /*
 617  * Note that we play around with the 'TS' bit in an attempt to get
 618  * the correct behaviour even in the presence of the asynchronous
 619  * IRQ13 behaviour
 620  */
 621 void math_error(void __user *ip)
 622 {
 623         struct task_struct *task;
 624         siginfo_t info;
 625         unsigned short cwd, swd, err;
 626
 627         /*
 628          * Save the info for the exception handler and clear the error.
 629          */
 630         task = current;
 631         save_init_fpu(task);
 632         task->thread.trap_no = 16;
 633         task->thread.error_code = 0;
 634         info.si_signo = SIGFPE;
 635         info.si_errno = 0;
 636         info.si_addr = ip;
 637         /*
 638          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 639          * status.  0x3f is the exception bits in these regs, 0x200 is the
 640          * C1 reg you need in case of a stack fault, 0x040 is the stack
 641          * fault bit.  We should only be taking one exception at a time,
 642          * so if this combination doesn't produce any single exception,
 643          * then we have a bad program that isn't synchronizing its FPU usage
 644          * and it will suffer the consequences since we won't be able to
 645          * fully reproduce the context of the exception
 646          */
 647         cwd = get_fpu_cwd(task);
 648         swd = get_fpu_swd(task);
 649
 650         err = swd & ~cwd;
 651
 652         if (err & 0x001) {      /* Invalid op */
 653                 /*
 654                  * swd & 0x240 == 0x040: Stack Underflow
 655                  * swd & 0x240 == 0x240: Stack Overflow
 656                  * User must clear the SF bit (0x40) if set
 657                  */
 658                 info.si_code = FPE_FLTINV;
 659         } else if (err & 0x004) { /* Divide by Zero */
 660                 info.si_code = FPE_FLTDIV;
 661         } else if (err & 0x008) { /* Overflow */
 662                 info.si_code = FPE_FLTOVF;
 663         } else if (err & 0x012) { /* Denormal, Underflow */
 664                 info.si_code = FPE_FLTUND;
 665         } else if (err & 0x020) { /* Precision */
 666                 info.si_code = FPE_FLTRES;
 667         } else {
 668                 /*
 669                  * If we're using IRQ 13, or supposedly even some trap 16
 670                  * implementations, it's possible we get a spurious trap...
 671                  */
 672                 return;         /* Spurious trap, no error */
 673         }
 674         force_sig_info(SIGFPE, &info, task);
 675 }
 676
 677 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
 678 {
 679         conditional_sti(regs);
 680
 681 #ifdef CONFIG_X86_32
 682         ignore_fpu_irq = 1;
 683 #else
 684         if (!user_mode(regs) &&
 685             kernel_math_error(regs, "kernel x87 math error", 16))
 686                 return;
 687 #endif
 688
 689         math_error((void __user *)regs->ip);
 690 }
 691
 692 static void simd_math_error(void __user *ip)
 693 {
 694         struct task_struct *task;
 695         siginfo_t info;
 696         unsigned short mxcsr;
 697
 698         /*
 699          * Save the info for the exception handler and clear the error.
 700          */
 701         task = current;
 702         save_init_fpu(task);
 703         task->thread.trap_no = 19;
 704         task->thread.error_code = 0;
 705         info.si_signo = SIGFPE;
 706         info.si_errno = 0;
 707         info.si_code = __SI_FAULT;
 708         info.si_addr = ip;
 709         /*
 710          * The SIMD FPU exceptions are handled a little differently, as there
 711          * is only a single status/control register.  Thus, to determine which
 712          * unmasked exception was caught we must mask the exception mask bits
 713          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 714          */
 715         mxcsr = get_fpu_mxcsr(task);
 716         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 717         case 0x000:
 718         default:
 719                 break;
 720         case 0x001: /* Invalid Op */
 721                 info.si_code = FPE_FLTINV;
 722                 break;
 723         case 0x002: /* Denormalize */
 724         case 0x010: /* Underflow */
 725                 info.si_code = FPE_FLTUND;
 726                 break;
 727         case 0x004: /* Zero Divide */
 728                 info.si_code = FPE_FLTDIV;
 729                 break;
 730         case 0x008: /* Overflow */
 731                 info.si_code = FPE_FLTOVF;
 732                 break;
 733         case 0x020: /* Precision */
 734                 info.si_code = FPE_FLTRES;
 735                 break;
 736         }
 737         force_sig_info(SIGFPE, &info, task);
 738 }
 739
 740 dotraplinkage void
 741 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
 742 {
 743         conditional_sti(regs);
 744
 745 #ifdef CONFIG_X86_32
 746         if (cpu_has_xmm) {
 747                 /* Handle SIMD FPU exceptions on PIII+ processors. */
 748                 ignore_fpu_irq = 1;
 749                 simd_math_error((void __user *)regs->ip);
 750                 return;
 751         }
 752         /*
 753          * Handle strange cache flush from user space exception
 754          * in all other cases.  This is undocumented behaviour.
 755          */
 756         if (regs->flags & X86_VM_MASK) {
 757                 handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code);
 758                 return;
 759         }
 760         current->thread.trap_no = 19;
 761         current->thread.error_code = error_code;
 762         die_if_kernel("cache flush denied", regs, error_code);
 763         force_sig(SIGSEGV, current);
 764 #else
 765         if (!user_mode(regs) &&
 766                         kernel_math_error(regs, "kernel simd math error", 19))
 767                 return;
 768         simd_math_error((void __user *)regs->ip);
 769 #endif
 770 }
 771
 772 dotraplinkage void
 773 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
 774 {
 775         conditional_sti(regs);
 776 #if 0
 777         /* No need to warn about this any longer. */
 778         printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
 779 #endif
 780 }
 781
 782 #ifdef CONFIG_X86_32
 783 unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp)
 784 {
 785         struct desc_struct *gdt = get_cpu_gdt_table(smp_processor_id());
 786         unsigned long base = (kesp - uesp) & -THREAD_SIZE;
 787         unsigned long new_kesp = kesp - base;
 788         unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
 789         __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
 790
 791         /* Set up base for espfix segment */
 792         desc &= 0x00f0ff0000000000ULL;
 793         desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
 794                 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
 795                 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
 796                 (lim_pages & 0xffff);
 797         *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
 798
 799         return new_kesp;
 800 }
 801 #else
 802 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 803 {
 804 }
 805
 806 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
 807 {
 808 }
 809 #endif
 810
 811 /*
 812  * 'math_state_restore()' saves the current math information in the
 813  * old math state array, and gets the new ones from the current task
 814  *
 815  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 816  * Don't touch unless you *really* know how it works.
 817  *
 818  * Must be called with kernel preemption disabled (in this case,
 819  * local interrupts are disabled at the call-site in entry.S).
 820  */
 821 asmlinkage void math_state_restore(void)
 822 {
 823         struct thread_info *thread = current_thread_info();
 824         struct task_struct *tsk = thread->task;
 825
 826         if (!tsk_used_math(tsk)) {
 827                 local_irq_enable();
 828                 /*
 829                  * does a slab alloc which can sleep
 830                  */
 831                 if (init_fpu(tsk)) {
 832                         /*
 833                          * ran out of memory!
 834                          */
 835                         do_group_exit(SIGKILL);
 836                         return;
 837                 }
 838                 local_irq_disable();
 839         }
 840
 841         clts();                         /* Allow maths ops (or we recurse) */
 842         /*
 843          * Paranoid restore. send a SIGSEGV if we fail to restore the state.
 844          */
 845         if (unlikely(restore_fpu_checking(tsk))) {
 846                 stts();
 847                 force_sig(SIGSEGV, tsk);
 848                 return;
 849         }
 850
 851         thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
 852         tsk->fpu_counter++;
 853 }
 854 EXPORT_SYMBOL_GPL(math_state_restore);
 855
 856 #ifndef CONFIG_MATH_EMULATION
 857 void math_emulate(struct math_emu_info *info)
 858 {
 859         printk(KERN_EMERG
 860                 "math-emulation not enabled and no coprocessor found.\n");
 861         printk(KERN_EMERG "killing %s.\n", current->comm);
 862         force_sig(SIGFPE, current);
 863         schedule();
 864 }
 865 #endif /* CONFIG_MATH_EMULATION */
 866
 867 dotraplinkage void __kprobes
 868 do_device_not_available(struct pt_regs *regs, long error_code)
 869 {
 870 #ifdef CONFIG_X86_32
 871         if (read_cr0() & X86_CR0_EM) {
 872                 struct math_emu_info info = { };
 873
 874                 conditional_sti(regs);
 875
 876                 info.regs = regs;
 877                 math_emulate(&info);
 878         } else {
 879                 math_state_restore(); /* interrupts still off */
 880                 conditional_sti(regs);
 881         }
 882 #else
 883         math_state_restore();
 884 #endif
 885 }
 886
 887 #ifdef CONFIG_X86_32
 888 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
 889 {
 890         siginfo_t info;
 891         local_irq_enable();
 892
 893         info.si_signo = SIGILL;
 894         info.si_errno = 0;
 895         info.si_code = ILL_BADSTK;
 896         info.si_addr = NULL;
 897         if (notify_die(DIE_TRAP, "iret exception",
 898                         regs, error_code, 32, SIGILL) == NOTIFY_STOP)
 899                 return;
 900         do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
 901 }
 902 #endif
 903
 904 void __init trap_init(void)
 905 {
 906         int i;
 907
 908 #ifdef CONFIG_EISA
 909         void __iomem *p = early_ioremap(0x0FFFD9, 4);
 910
 911         if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
 912                 EISA_bus = 1;
 913         early_iounmap(p, 4);
 914 #endif
 915
 916         set_intr_gate(0, &divide_error);
 917         set_intr_gate_ist(1, &debug, DEBUG_STACK);
 918         set_intr_gate_ist(2, &nmi, NMI_STACK);
 919         /* int3 can be called from all */
 920         set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
 921         /* int4 can be called from all */
 922         set_system_intr_gate(4, &overflow);
 923         set_intr_gate(5, &bounds);
 924         set_intr_gate(6, &invalid_op);
 925         set_intr_gate(7, &device_not_available);
 926 #ifdef CONFIG_X86_32
 927         set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
 928 #else
 929         set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
 930 #endif
 931         set_intr_gate(9, &coprocessor_segment_overrun);
 932         set_intr_gate(10, &invalid_TSS);
 933         set_intr_gate(11, &segment_not_present);
 934         set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
 935         set_intr_gate(13, &general_protection);
 936         set_intr_gate(14, &page_fault);
 937         set_intr_gate(15, &spurious_interrupt_bug);
 938         set_intr_gate(16, &coprocessor_error);
 939         set_intr_gate(17, &alignment_check);
 940 #ifdef CONFIG_X86_MCE
 941         set_intr_gate_ist(18, &machine_check, MCE_STACK);
 942 #endif
 943         set_intr_gate(19, &simd_coprocessor_error);
 944
 945 #ifdef CONFIG_IA32_EMULATION
 946         set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 947 #endif
 948
 949 #ifdef CONFIG_X86_32
 950         if (cpu_has_fxsr) {
 951                 printk(KERN_INFO "Enabling fast FPU save and restore... ");
 952                 set_in_cr4(X86_CR4_OSFXSR);
 953                 printk("done.\n");
 954         }
 955         if (cpu_has_xmm) {
 956                 printk(KERN_INFO
 957                         "Enabling unmasked SIMD FPU exception support... ");
 958                 set_in_cr4(X86_CR4_OSXMMEXCPT);
 959                 printk("done.\n");
 960         }
 961
 962         set_system_trap_gate(SYSCALL_VECTOR, &system_call);
 963 #endif
 964
 965         /* Reserve all the builtin and the syscall vector: */
 966         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
 967                 set_bit(i, used_vectors);
 968
 969 #ifdef CONFIG_X86_64
 970         set_bit(IA32_SYSCALL_VECTOR, used_vectors);
 971 #else
 972         set_bit(SYSCALL_VECTOR, used_vectors);
 973 #endif
 974         /*
 975          * Should be a barrier for any external CPU state:
 976          */
 977         cpu_init();
 978
 979 #ifdef CONFIG_X86_32
 980         x86_quirk_trap_init();
 981 #endif
 982 }