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
  58 #include <asm/mach_traps.h>
  59
  60 #ifdef CONFIG_X86_64
  61 #include <asm/pgalloc.h>
  62 #include <asm/proto.h>
  63 #else
  64 #include <asm/processor-flags.h>
  65 #include <asm/setup.h>
  66 #include <asm/traps.h>
  67
  68 #include "cpu/mcheck/mce.h"
  69
  70 asmlinkage int system_call(void);
  71
  72 /* Do we ignore FPU interrupts ? */
  73 char ignore_fpu_irq;
  74
  75 /*
  76  * The IDT has to be page-aligned to simplify the Pentium
  77  * F0 0F bug workaround.. We have a special link segment
  78  * for this.
  79  */
  80 gate_desc idt_table[256]
  81         __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
  82 #endif
  83
  84 DECLARE_BITMAP(used_vectors, NR_VECTORS);
  85 EXPORT_SYMBOL_GPL(used_vectors);
  86
  87 static int ignore_nmis;
  88
  89 static inline void conditional_sti(struct pt_regs *regs)
  90 {
  91         if (regs->flags & X86_EFLAGS_IF)
  92                 local_irq_enable();
  93 }
  94
  95 static inline void preempt_conditional_sti(struct pt_regs *regs)
  96 {
  97         inc_preempt_count();
  98         if (regs->flags & X86_EFLAGS_IF)
  99                 local_irq_enable();
 100 }
 101
 102 static inline void conditional_cli(struct pt_regs *regs)
 103 {
 104         if (regs->flags & X86_EFLAGS_IF)
 105                 local_irq_disable();
 106 }
 107
 108 static inline void preempt_conditional_cli(struct pt_regs *regs)
 109 {
 110         if (regs->flags & X86_EFLAGS_IF)
 111                 local_irq_disable();
 112         dec_preempt_count();
 113 }
 114
 115 #ifdef CONFIG_X86_32
 116 static inline void
 117 die_if_kernel(const char *str, struct pt_regs *regs, long err)
 118 {
 119         if (!user_mode_vm(regs))
 120                 die(str, regs, err);
 121 }
 122 #endif
 123
 124 static void __kprobes
 125 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
 126         long error_code, siginfo_t *info)
 127 {
 128         struct task_struct *tsk = current;
 129
 130 #ifdef CONFIG_X86_32
 131         if (regs->flags & X86_VM_MASK) {
 132                 /*
 133                  * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
 134                  * On nmi (interrupt 2), do_trap should not be called.
 135                  */
 136                 if (trapnr < 6)
 137                         goto vm86_trap;
 138                 goto trap_signal;
 139         }
 140 #endif
 141
 142         if (!user_mode(regs))
 143                 goto kernel_trap;
 144
 145 #ifdef CONFIG_X86_32
 146 trap_signal:
 147 #endif
 148         /*
 149          * We want error_code and trap_no set for userspace faults and
 150          * kernelspace faults which result in die(), but not
 151          * kernelspace faults which are fixed up.  die() gives the
 152          * process no chance to handle the signal and notice the
 153          * kernel fault information, so that won't result in polluting
 154          * the information about previously queued, but not yet
 155          * delivered, faults.  See also do_general_protection below.
 156          */
 157         tsk->thread.error_code = error_code;
 158         tsk->thread.trap_no = trapnr;
 159
 160 #ifdef CONFIG_X86_64
 161         if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
 162             printk_ratelimit()) {
 163                 printk(KERN_INFO
 164                        "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
 165                        tsk->comm, tsk->pid, str,
 166                        regs->ip, regs->sp, error_code);
 167                 print_vma_addr(" in ", regs->ip);
 168                 printk("\n");
 169         }
 170 #endif
 171
 172         if (info)
 173                 force_sig_info(signr, info, tsk);
 174         else
 175                 force_sig(signr, tsk);
 176         return;
 177
 178 kernel_trap:
 179         if (!fixup_exception(regs)) {
 180                 tsk->thread.error_code = error_code;
 181                 tsk->thread.trap_no = trapnr;
 182                 die(str, regs, error_code);
 183         }
 184         return;
 185
 186 #ifdef CONFIG_X86_32
 187 vm86_trap:
 188         if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
 189                                                 error_code, trapnr))
 190                 goto trap_signal;
 191         return;
 192 #endif
 193 }
 194
 195 #define DO_ERROR(trapnr, signr, str, name)                              \
 196 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 197 {                                                                       \
 198         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 199                                                         == NOTIFY_STOP) \
 200                 return;                                                 \
 201         conditional_sti(regs);                                          \
 202         do_trap(trapnr, signr, str, regs, error_code, NULL);            \
 203 }
 204
 205 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr)         \
 206 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 207 {                                                                       \
 208         siginfo_t info;                                                 \
 209         info.si_signo = signr;                                          \
 210         info.si_errno = 0;                                              \
 211         info.si_code = sicode;                                          \
 212         info.si_addr = (void __user *)siaddr;                           \
 213         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 214                                                         == NOTIFY_STOP) \
 215                 return;                                                 \
 216         conditional_sti(regs);                                          \
 217         do_trap(trapnr, signr, str, regs, error_code, &info);           \
 218 }
 219
 220 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
 221 DO_ERROR(4, SIGSEGV, "overflow", overflow)
 222 DO_ERROR(5, SIGSEGV, "bounds", bounds)
 223 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
 224 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
 225 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 226 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
 227 #ifdef CONFIG_X86_32
 228 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
 229 #endif
 230 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 231
 232 #ifdef CONFIG_X86_64
 233 /* Runs on IST stack */
 234 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 235 {
 236         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 237                         12, SIGBUS) == NOTIFY_STOP)
 238                 return;
 239         preempt_conditional_sti(regs);
 240         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 241         preempt_conditional_cli(regs);
 242 }
 243
 244 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
 245 {
 246         static const char str[] = "double fault";
 247         struct task_struct *tsk = current;
 248
 249         /* Return not checked because double check cannot be ignored */
 250         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 251
 252         tsk->thread.error_code = error_code;
 253         tsk->thread.trap_no = 8;
 254
 255         /*
 256          * This is always a kernel trap and never fixable (and thus must
 257          * never return).
 258          */
 259         for (;;)
 260                 die(str, regs, error_code);
 261 }
 262 #endif
 263
 264 dotraplinkage void __kprobes
 265 do_general_protection(struct pt_regs *regs, long error_code)
 266 {
 267         struct task_struct *tsk;
 268
 269         conditional_sti(regs);
 270
 271 #ifdef CONFIG_X86_32
 272         if (regs->flags & X86_VM_MASK)
 273                 goto gp_in_vm86;
 274 #endif
 275
 276         tsk = current;
 277         if (!user_mode(regs))
 278                 goto gp_in_kernel;
 279
 280         tsk->thread.error_code = error_code;
 281         tsk->thread.trap_no = 13;
 282
 283         if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
 284                         printk_ratelimit()) {
 285                 printk(KERN_INFO
 286                         "%s[%d] general protection ip:%lx sp:%lx error:%lx",
 287                         tsk->comm, task_pid_nr(tsk),
 288                         regs->ip, regs->sp, error_code);
 289                 print_vma_addr(" in ", regs->ip);
 290                 printk("\n");
 291         }
 292
 293         force_sig(SIGSEGV, tsk);
 294         return;
 295
 296 #ifdef CONFIG_X86_32
 297 gp_in_vm86:
 298         local_irq_enable();
 299         handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 300         return;
 301 #endif
 302
 303 gp_in_kernel:
 304         if (fixup_exception(regs))
 305                 return;
 306
 307         tsk->thread.error_code = error_code;
 308         tsk->thread.trap_no = 13;
 309         if (notify_die(DIE_GPF, "general protection fault", regs,
 310                                 error_code, 13, SIGSEGV) == NOTIFY_STOP)
 311                 return;
 312         die("general protection fault", regs, error_code);
 313 }
 314
 315 static notrace __kprobes void
 316 mem_parity_error(unsigned char reason, struct pt_regs *regs)
 317 {
 318         printk(KERN_EMERG
 319                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 320                         reason, smp_processor_id());
 321
 322         printk(KERN_EMERG
 323                 "You have some hardware problem, likely on the PCI bus.\n");
 324
 325 #if defined(CONFIG_EDAC)
 326         if (edac_handler_set()) {
 327                 edac_atomic_assert_error();
 328                 return;
 329         }
 330 #endif
 331
 332         if (panic_on_unrecovered_nmi)
 333                 panic("NMI: Not continuing");
 334
 335         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 336
 337         /* Clear and disable the memory parity error line. */
 338         reason = (reason & 0xf) | 4;
 339         outb(reason, 0x61);
 340 }
 341
 342 static notrace __kprobes void
 343 io_check_error(unsigned char reason, struct pt_regs *regs)
 344 {
 345         unsigned long i;
 346
 347         printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
 348         show_registers(regs);
 349
 350         /* Re-enable the IOCK line, wait for a few seconds */
 351         reason = (reason & 0xf) | 8;
 352         outb(reason, 0x61);
 353
 354         i = 2000;
 355         while (--i)
 356                 udelay(1000);
 357
 358         reason &= ~8;
 359         outb(reason, 0x61);
 360 }
 361
 362 static notrace __kprobes void
 363 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
 364 {
 365         if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
 366                         NOTIFY_STOP)
 367                 return;
 368 #ifdef CONFIG_MCA
 369         /*
 370          * Might actually be able to figure out what the guilty party
 371          * is:
 372          */
 373         if (MCA_bus) {
 374                 mca_handle_nmi();
 375                 return;
 376         }
 377 #endif
 378         printk(KERN_EMERG
 379                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 380                         reason, smp_processor_id());
 381
 382         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 383         if (panic_on_unrecovered_nmi)
 384                 panic("NMI: Not continuing");
 385
 386         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 387 }
 388
 389 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
 390 {
 391         unsigned char reason = 0;
 392         int cpu;
 393
 394         cpu = smp_processor_id();
 395
 396         /* Only the BSP gets external NMIs from the system. */
 397         if (!cpu)
 398                 reason = get_nmi_reason();
 399
 400         if (!(reason & 0xc0)) {
 401                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 402                                                                 == NOTIFY_STOP)
 403                         return;
 404 #ifdef CONFIG_X86_LOCAL_APIC
 405                 /*
 406                  * Ok, so this is none of the documented NMI sources,
 407                  * so it must be the NMI watchdog.
 408                  */
 409                 if (nmi_watchdog_tick(regs, reason))
 410                         return;
 411                 if (!do_nmi_callback(regs, cpu))
 412                         unknown_nmi_error(reason, regs);
 413 #else
 414                 unknown_nmi_error(reason, regs);
 415 #endif
 416
 417                 return;
 418         }
 419         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 420                 return;
 421
 422         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 423         if (reason & 0x80)
 424                 mem_parity_error(reason, regs);
 425         if (reason & 0x40)
 426                 io_check_error(reason, regs);
 427 #ifdef CONFIG_X86_32
 428         /*
 429          * Reassert NMI in case it became active meanwhile
 430          * as it's edge-triggered:
 431          */
 432         reassert_nmi();
 433 #endif
 434 }
 435
 436 dotraplinkage notrace __kprobes void
 437 do_nmi(struct pt_regs *regs, long error_code)
 438 {
 439         nmi_enter();
 440
 441         inc_irq_stat(__nmi_count);
 442
 443         if (!ignore_nmis)
 444                 default_do_nmi(regs);
 445
 446         nmi_exit();
 447 }
 448
 449 void stop_nmi(void)
 450 {
 451         acpi_nmi_disable();
 452         ignore_nmis++;
 453 }
 454
 455 void restart_nmi(void)
 456 {
 457         ignore_nmis--;
 458         acpi_nmi_enable();
 459 }
 460
 461 /* May run on IST stack. */
 462 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
 463 {
 464 #ifdef CONFIG_KPROBES
 465         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 466                         == NOTIFY_STOP)
 467                 return;
 468 #else
 469         if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
 470                         == NOTIFY_STOP)
 471                 return;
 472 #endif
 473
 474         preempt_conditional_sti(regs);
 475         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 476         preempt_conditional_cli(regs);
 477 }
 478
 479 #ifdef CONFIG_X86_64
 480 /*
 481  * Help handler running on IST stack to switch back to user stack
 482  * for scheduling or signal handling. The actual stack switch is done in
 483  * entry.S
 484  */
 485 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 486 {
 487         struct pt_regs *regs = eregs;
 488         /* Did already sync */
 489         if (eregs == (struct pt_regs *)eregs->sp)
 490                 ;
 491         /* Exception from user space */
 492         else if (user_mode(eregs))
 493                 regs = task_pt_regs(current);
 494         /*
 495          * Exception from kernel and interrupts are enabled. Move to
 496          * kernel process stack.
 497          */
 498         else if (eregs->flags & X86_EFLAGS_IF)
 499                 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
 500         if (eregs != regs)
 501                 *regs = *eregs;
 502         return regs;
 503 }
 504 #endif
 505
 506 /*
 507  * Our handling of the processor debug registers is non-trivial.
 508  * We do not clear them on entry and exit from the kernel. Therefore
 509  * it is possible to get a watchpoint trap here from inside the kernel.
 510  * However, the code in ./ptrace.c has ensured that the user can
 511  * only set watchpoints on userspace addresses. Therefore the in-kernel
 512  * watchpoint trap can only occur in code which is reading/writing
 513  * from user space. Such code must not hold kernel locks (since it
 514  * can equally take a page fault), therefore it is safe to call
 515  * force_sig_info even though that claims and releases locks.
 516  *
 517  * Code in ./signal.c ensures that the debug control register
 518  * is restored before we deliver any signal, and therefore that
 519  * user code runs with the correct debug control register even though
 520  * we clear it here.
 521  *
 522  * Being careful here means that we don't have to be as careful in a
 523  * lot of more complicated places (task switching can be a bit lazy
 524  * about restoring all the debug state, and ptrace doesn't have to
 525  * find every occurrence of the TF bit that could be saved away even
 526  * by user code)
 527  *
 528  * May run on IST stack.
 529  */
 530 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
 531 {
 532         struct task_struct *tsk = current;
 533         unsigned long condition;
 534         int si_code;
 535
 536         get_debugreg(condition, 6);
 537
 538         /* Catch kmemcheck conditions first of all! */
 539         if (condition & DR_STEP && kmemcheck_trap(regs))
 540                 return;
 541
 542         /*
 543          * The processor cleared BTF, so don't mark that we need it set.
 544          */
 545         clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
 546         tsk->thread.debugctlmsr = 0;
 547
 548         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 549                                                 SIGTRAP) == NOTIFY_STOP)
 550                 return;
 551
 552         /* It's safe to allow irq's after DR6 has been saved */
 553         preempt_conditional_sti(regs);
 554
 555         /* Mask out spurious debug traps due to lazy DR7 setting */
 556         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 557                 if (!tsk->thread.debugreg7)
 558                         goto clear_dr7;
 559         }
 560
 561 #ifdef CONFIG_X86_32
 562         if (regs->flags & X86_VM_MASK)
 563                 goto debug_vm86;
 564 #endif
 565
 566         /* Save debug status register where ptrace can see it */
 567         tsk->thread.debugreg6 = condition;
 568
 569         /*
 570          * Single-stepping through TF: make sure we ignore any events in
 571          * kernel space (but re-enable TF when returning to user mode).
 572          */
 573         if (condition & DR_STEP) {
 574                 if (!user_mode(regs))
 575                         goto clear_TF_reenable;
 576         }
 577
 578         si_code = get_si_code(condition);
 579         /* Ok, finally something we can handle */
 580         send_sigtrap(tsk, regs, error_code, si_code);
 581
 582         /*
 583          * Disable additional traps. They'll be re-enabled when
 584          * the signal is delivered.
 585          */
 586 clear_dr7:
 587         set_debugreg(0, 7);
 588         preempt_conditional_cli(regs);
 589         return;
 590
 591 #ifdef CONFIG_X86_32
 592 debug_vm86:
 593         /* reenable preemption: handle_vm86_trap() might sleep */
 594         dec_preempt_count();
 595         handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
 596         conditional_cli(regs);
 597         return;
 598 #endif
 599
 600 clear_TF_reenable:
 601         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 602         regs->flags &= ~X86_EFLAGS_TF;
 603         preempt_conditional_cli(regs);
 604         return;
 605 }
 606
 607 #ifdef CONFIG_X86_64
 608 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 609 {
 610         if (fixup_exception(regs))
 611                 return 1;
 612
 613         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 614         /* Illegal floating point operation in the kernel */
 615         current->thread.trap_no = trapnr;
 616         die(str, regs, 0);
 617         return 0;
 618 }
 619 #endif
 620
 621 /*
 622  * Note that we play around with the 'TS' bit in an attempt to get
 623  * the correct behaviour even in the presence of the asynchronous
 624  * IRQ13 behaviour
 625  */
 626 void math_error(void __user *ip)
 627 {
 628         struct task_struct *task;
 629         siginfo_t info;
 630         unsigned short cwd, swd, err;
 631
 632         /*
 633          * Save the info for the exception handler and clear the error.
 634          */
 635         task = current;
 636         save_init_fpu(task);
 637         task->thread.trap_no = 16;
 638         task->thread.error_code = 0;
 639         info.si_signo = SIGFPE;
 640         info.si_errno = 0;
 641         info.si_addr = ip;
 642         /*
 643          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 644          * status.  0x3f is the exception bits in these regs, 0x200 is the
 645          * C1 reg you need in case of a stack fault, 0x040 is the stack
 646          * fault bit.  We should only be taking one exception at a time,
 647          * so if this combination doesn't produce any single exception,
 648          * then we have a bad program that isn't synchronizing its FPU usage
 649          * and it will suffer the consequences since we won't be able to
 650          * fully reproduce the context of the exception
 651          */
 652         cwd = get_fpu_cwd(task);
 653         swd = get_fpu_swd(task);
 654
 655         err = swd & ~cwd;
 656
 657         if (err & 0x001) {      /* Invalid op */
 658                 /*
 659                  * swd & 0x240 == 0x040: Stack Underflow
 660                  * swd & 0x240 == 0x240: Stack Overflow
 661                  * User must clear the SF bit (0x40) if set
 662                  */
 663                 info.si_code = FPE_FLTINV;
 664         } else if (err & 0x004) { /* Divide by Zero */
 665                 info.si_code = FPE_FLTDIV;
 666         } else if (err & 0x008) { /* Overflow */
 667                 info.si_code = FPE_FLTOVF;
 668         } else if (err & 0x012) { /* Denormal, Underflow */
 669                 info.si_code = FPE_FLTUND;
 670         } else if (err & 0x020) { /* Precision */
 671                 info.si_code = FPE_FLTRES;
 672         } else {
 673                 /*
 674                  * If we're using IRQ 13, or supposedly even some trap 16
 675                  * implementations, it's possible we get a spurious trap...
 676                  */
 677                 return;         /* Spurious trap, no error */
 678         }
 679         force_sig_info(SIGFPE, &info, task);
 680 }
 681
 682 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
 683 {
 684         conditional_sti(regs);
 685
 686 #ifdef CONFIG_X86_32
 687         ignore_fpu_irq = 1;
 688 #else
 689         if (!user_mode(regs) &&
 690             kernel_math_error(regs, "kernel x87 math error", 16))
 691                 return;
 692 #endif
 693
 694         math_error((void __user *)regs->ip);
 695 }
 696
 697 static void simd_math_error(void __user *ip)
 698 {
 699         struct task_struct *task;
 700         siginfo_t info;
 701         unsigned short mxcsr;
 702
 703         /*
 704          * Save the info for the exception handler and clear the error.
 705          */
 706         task = current;
 707         save_init_fpu(task);
 708         task->thread.trap_no = 19;
 709         task->thread.error_code = 0;
 710         info.si_signo = SIGFPE;
 711         info.si_errno = 0;
 712         info.si_code = __SI_FAULT;
 713         info.si_addr = ip;
 714         /*
 715          * The SIMD FPU exceptions are handled a little differently, as there
 716          * is only a single status/control register.  Thus, to determine which
 717          * unmasked exception was caught we must mask the exception mask bits
 718          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 719          */
 720         mxcsr = get_fpu_mxcsr(task);
 721         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 722         case 0x000:
 723         default:
 724                 break;
 725         case 0x001: /* Invalid Op */
 726                 info.si_code = FPE_FLTINV;
 727                 break;
 728         case 0x002: /* Denormalize */
 729         case 0x010: /* Underflow */
 730                 info.si_code = FPE_FLTUND;
 731                 break;
 732         case 0x004: /* Zero Divide */
 733                 info.si_code = FPE_FLTDIV;
 734                 break;
 735         case 0x008: /* Overflow */
 736                 info.si_code = FPE_FLTOVF;
 737                 break;
 738         case 0x020: /* Precision */
 739                 info.si_code = FPE_FLTRES;
 740                 break;
 741         }
 742         force_sig_info(SIGFPE, &info, task);
 743 }
 744
 745 dotraplinkage void
 746 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
 747 {
 748         conditional_sti(regs);
 749
 750 #ifdef CONFIG_X86_32
 751         if (cpu_has_xmm) {
 752                 /* Handle SIMD FPU exceptions on PIII+ processors. */
 753                 ignore_fpu_irq = 1;
 754                 simd_math_error((void __user *)regs->ip);
 755                 return;
 756         }
 757         /*
 758          * Handle strange cache flush from user space exception
 759          * in all other cases.  This is undocumented behaviour.
 760          */
 761         if (regs->flags & X86_VM_MASK) {
 762                 handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code);
 763                 return;
 764         }
 765         current->thread.trap_no = 19;
 766         current->thread.error_code = error_code;
 767         die_if_kernel("cache flush denied", regs, error_code);
 768         force_sig(SIGSEGV, current);
 769 #else
 770         if (!user_mode(regs) &&
 771                         kernel_math_error(regs, "kernel simd math error", 19))
 772                 return;
 773         simd_math_error((void __user *)regs->ip);
 774 #endif
 775 }
 776
 777 dotraplinkage void
 778 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
 779 {
 780         conditional_sti(regs);
 781 #if 0
 782         /* No need to warn about this any longer. */
 783         printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
 784 #endif
 785 }
 786
 787 #ifdef CONFIG_X86_32
 788 unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp)
 789 {
 790         struct desc_struct *gdt = get_cpu_gdt_table(smp_processor_id());
 791         unsigned long base = (kesp - uesp) & -THREAD_SIZE;
 792         unsigned long new_kesp = kesp - base;
 793         unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
 794         __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
 795
 796         /* Set up base for espfix segment */
 797         desc &= 0x00f0ff0000000000ULL;
 798         desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
 799                 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
 800                 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
 801                 (lim_pages & 0xffff);
 802         *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
 803
 804         return new_kesp;
 805 }
 806 #endif
 807
 808 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 809 {
 810 }
 811
 812 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
 813 {
 814 }
 815
 816 /*
 817  * 'math_state_restore()' saves the current math information in the
 818  * old math state array, and gets the new ones from the current task
 819  *
 820  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 821  * Don't touch unless you *really* know how it works.
 822  *
 823  * Must be called with kernel preemption disabled (in this case,
 824  * local interrupts are disabled at the call-site in entry.S).
 825  */
 826 asmlinkage void math_state_restore(void)
 827 {
 828         struct thread_info *thread = current_thread_info();
 829         struct task_struct *tsk = thread->task;
 830
 831         if (!tsk_used_math(tsk)) {
 832                 local_irq_enable();
 833                 /*
 834                  * does a slab alloc which can sleep
 835                  */
 836                 if (init_fpu(tsk)) {
 837                         /*
 838                          * ran out of memory!
 839                          */
 840                         do_group_exit(SIGKILL);
 841                         return;
 842                 }
 843                 local_irq_disable();
 844         }
 845
 846         clts();                         /* Allow maths ops (or we recurse) */
 847         /*
 848          * Paranoid restore. send a SIGSEGV if we fail to restore the state.
 849          */
 850         if (unlikely(restore_fpu_checking(tsk))) {
 851                 stts();
 852                 force_sig(SIGSEGV, tsk);
 853                 return;
 854         }
 855
 856         thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
 857         tsk->fpu_counter++;
 858 }
 859 EXPORT_SYMBOL_GPL(math_state_restore);
 860
 861 #ifndef CONFIG_MATH_EMULATION
 862 void math_emulate(struct math_emu_info *info)
 863 {
 864         printk(KERN_EMERG
 865                 "math-emulation not enabled and no coprocessor found.\n");
 866         printk(KERN_EMERG "killing %s.\n", current->comm);
 867         force_sig(SIGFPE, current);
 868         schedule();
 869 }
 870 #endif /* CONFIG_MATH_EMULATION */
 871
 872 dotraplinkage void __kprobes
 873 do_device_not_available(struct pt_regs *regs, long error_code)
 874 {
 875 #ifdef CONFIG_X86_32
 876         if (read_cr0() & X86_CR0_EM) {
 877                 struct math_emu_info info = { };
 878
 879                 conditional_sti(regs);
 880
 881                 info.regs = regs;
 882                 math_emulate(&info);
 883         } else {
 884                 math_state_restore(); /* interrupts still off */
 885                 conditional_sti(regs);
 886         }
 887 #else
 888         math_state_restore();
 889 #endif
 890 }
 891
 892 #ifdef CONFIG_X86_32
 893 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
 894 {
 895         siginfo_t info;
 896         local_irq_enable();
 897
 898         info.si_signo = SIGILL;
 899         info.si_errno = 0;
 900         info.si_code = ILL_BADSTK;
 901         info.si_addr = NULL;
 902         if (notify_die(DIE_TRAP, "iret exception",
 903                         regs, error_code, 32, SIGILL) == NOTIFY_STOP)
 904                 return;
 905         do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
 906 }
 907 #endif
 908
 909 void __init trap_init(void)
 910 {
 911         int i;
 912
 913 #ifdef CONFIG_EISA
 914         void __iomem *p = early_ioremap(0x0FFFD9, 4);
 915
 916         if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
 917                 EISA_bus = 1;
 918         early_iounmap(p, 4);
 919 #endif
 920
 921         set_intr_gate(0, &divide_error);
 922         set_intr_gate_ist(1, &debug, DEBUG_STACK);
 923         set_intr_gate_ist(2, &nmi, NMI_STACK);
 924         /* int3 can be called from all */
 925         set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
 926         /* int4 can be called from all */
 927         set_system_intr_gate(4, &overflow);
 928         set_intr_gate(5, &bounds);
 929         set_intr_gate(6, &invalid_op);
 930         set_intr_gate(7, &device_not_available);
 931 #ifdef CONFIG_X86_32
 932         set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
 933 #else
 934         set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
 935 #endif
 936         set_intr_gate(9, &coprocessor_segment_overrun);
 937         set_intr_gate(10, &invalid_TSS);
 938         set_intr_gate(11, &segment_not_present);
 939         set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
 940         set_intr_gate(13, &general_protection);
 941         set_intr_gate(14, &page_fault);
 942         set_intr_gate(15, &spurious_interrupt_bug);
 943         set_intr_gate(16, &coprocessor_error);
 944         set_intr_gate(17, &alignment_check);
 945 #ifdef CONFIG_X86_MCE
 946         set_intr_gate_ist(18, &machine_check, MCE_STACK);
 947 #endif
 948         set_intr_gate(19, &simd_coprocessor_error);
 949
 950         /* Reserve all the builtin and the syscall vector: */
 951         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
 952                 set_bit(i, used_vectors);
 953
 954 #ifdef CONFIG_IA32_EMULATION
 955         set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 956         set_bit(IA32_SYSCALL_VECTOR, used_vectors);
 957 #endif
 958
 959 #ifdef CONFIG_X86_32
 960         if (cpu_has_fxsr) {
 961                 printk(KERN_INFO "Enabling fast FPU save and restore... ");
 962                 set_in_cr4(X86_CR4_OSFXSR);
 963                 printk("done.\n");
 964         }
 965         if (cpu_has_xmm) {
 966                 printk(KERN_INFO
 967                         "Enabling unmasked SIMD FPU exception support... ");
 968                 set_in_cr4(X86_CR4_OSXMMEXCPT);
 969                 printk("done.\n");
 970         }
 971
 972         set_system_trap_gate(SYSCALL_VECTOR, &system_call);
 973         set_bit(SYSCALL_VECTOR, used_vectors);
 974 #endif
 975
 976         /*
 977          * Should be a barrier for any external CPU state:
 978          */
 979         cpu_init();
 980
 981 #ifdef CONFIG_X86_32
 982         x86_quirk_trap_init();
 983 #endif
 984 }