sys/i386/i386/trap.c

   1 /*-
   2  * Copyright (C) 1994, David Greenman
   3  * Copyright (c) 1990, 1993
   4  *      The Regents of the University of California.  All rights reserved.
   5  *
   6  * This code is derived from software contributed to Berkeley by
   7  * the University of Utah, and William Jolitz.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in the
  16  *    documentation and/or other materials provided with the distribution.
  17  * 3. All advertising materials mentioning features or use of this software
  18  *    must display the following acknowledgement:
  19  *      This product includes software developed by the University of
  20  *      California, Berkeley and its contributors.
  21  * 4. Neither the name of the University nor the names of its contributors
  22  *    may be used to endorse or promote products derived from this software
  23  *    without specific prior written permission.
  24  *
  25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  35  * SUCH DAMAGE.
  36  *
  37  *      from: @(#)trap.c        7.4 (Berkeley) 5/13/91
  38  */
  39
  40 #include <sys/cdefs.h>
  41 __FBSDID("$FreeBSD$");
  42
  43 /*
  44  * 386 Trap and System call handling
  45  */
  46
  47 #include "opt_clock.h"
  48 #include "opt_cpu.h"
  49 #include "opt_hwpmc_hooks.h"
  50 #include "opt_isa.h"
  51 #include "opt_kdb.h"
  52 #include "opt_npx.h"
  53 #include "opt_stack.h"
  54 #include "opt_trap.h"
  55
  56 #include <sys/param.h>
  57 #include <sys/bus.h>
  58 #include <sys/systm.h>
  59 #include <sys/proc.h>
  60 #include <sys/pioctl.h>
  61 #include <sys/ptrace.h>
  62 #include <sys/kdb.h>
  63 #include <sys/kernel.h>
  64 #include <sys/ktr.h>
  65 #include <sys/lock.h>
  66 #include <sys/mutex.h>
  67 #include <sys/resourcevar.h>
  68 #include <sys/signalvar.h>
  69 #include <sys/syscall.h>
  70 #include <sys/sysctl.h>
  71 #include <sys/sysent.h>
  72 #include <sys/uio.h>
  73 #include <sys/vmmeter.h>
  74 #ifdef HWPMC_HOOKS
  75 #include <sys/pmckern.h>
  76 PMC_SOFT_DEFINE( , , page_fault, all);
  77 PMC_SOFT_DEFINE( , , page_fault, read);
  78 PMC_SOFT_DEFINE( , , page_fault, write);
  79 #endif
  80 #include <security/audit/audit.h>
  81
  82 #include <vm/vm.h>
  83 #include <vm/vm_param.h>
  84 #include <vm/pmap.h>
  85 #include <vm/vm_kern.h>
  86 #include <vm/vm_map.h>
  87 #include <vm/vm_page.h>
  88 #include <vm/vm_extern.h>
  89
  90 #include <machine/cpu.h>
  91 #include <machine/intr_machdep.h>
  92 #include <x86/mca.h>
  93 #include <machine/md_var.h>
  94 #include <machine/pcb.h>
  95 #ifdef SMP
  96 #include <machine/smp.h>
  97 #endif
  98 #include <machine/stack.h>
  99 #include <machine/tss.h>
 100 #include <machine/vm86.h>
 101
 102 #ifdef POWERFAIL_NMI
 103 #include <sys/syslog.h>
 104 #include <machine/clock.h>
 105 #endif
 106
 107 #ifdef KDTRACE_HOOKS
 108 #include <sys/dtrace_bsd.h>
 109 #endif
 110
 111 extern void trap(struct trapframe *frame);
 112 extern void syscall(struct trapframe *frame);
 113
 114 static int trap_pfault(struct trapframe *, int, vm_offset_t);
 115 static void trap_fatal(struct trapframe *, vm_offset_t);
 116 void dblfault_handler(void);
 117
 118 extern inthand_t IDTVEC(lcall_syscall);
 119
 120 #define MAX_TRAP_MSG            32
 121 static char *trap_msg[] = {
 122         "",                                     /*  0 unused */
 123         "privileged instruction fault",         /*  1 T_PRIVINFLT */
 124         "",                                     /*  2 unused */
 125         "breakpoint instruction fault",         /*  3 T_BPTFLT */
 126         "",                                     /*  4 unused */
 127         "",                                     /*  5 unused */
 128         "arithmetic trap",                      /*  6 T_ARITHTRAP */
 129         "",                                     /*  7 unused */
 130         "",                                     /*  8 unused */
 131         "general protection fault",             /*  9 T_PROTFLT */
 132         "trace trap",                           /* 10 T_TRCTRAP */
 133         "",                                     /* 11 unused */
 134         "page fault",                           /* 12 T_PAGEFLT */
 135         "",                                     /* 13 unused */
 136         "alignment fault",                      /* 14 T_ALIGNFLT */
 137         "",                                     /* 15 unused */
 138         "",                                     /* 16 unused */
 139         "",                                     /* 17 unused */
 140         "integer divide fault",                 /* 18 T_DIVIDE */
 141         "non-maskable interrupt trap",          /* 19 T_NMI */
 142         "overflow trap",                        /* 20 T_OFLOW */
 143         "FPU bounds check fault",               /* 21 T_BOUND */
 144         "FPU device not available",             /* 22 T_DNA */
 145         "double fault",                         /* 23 T_DOUBLEFLT */
 146         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
 147         "invalid TSS fault",                    /* 25 T_TSSFLT */
 148         "segment not present fault",            /* 26 T_SEGNPFLT */
 149         "stack fault",                          /* 27 T_STKFLT */
 150         "machine check trap",                   /* 28 T_MCHK */
 151         "SIMD floating-point exception",        /* 29 T_XMMFLT */
 152         "reserved (unknown) fault",             /* 30 T_RESERVED */
 153         "",                                     /* 31 unused (reserved) */
 154         "DTrace pid return trap",               /* 32 T_DTRACE_RET */
 155 };
 156
 157 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 158 int has_f00f_bug = 0;           /* Initialized so that it can be patched. */
 159 #endif
 160
 161 #ifdef KDB
 162 static int kdb_on_nmi = 1;
 163 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RWTUN,
 164         &kdb_on_nmi, 0, "Go to KDB on NMI");
 165 #endif
 166 static int panic_on_nmi = 1;
 167 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RWTUN,
 168         &panic_on_nmi, 0, "Panic on NMI");
 169 static int prot_fault_translation = 0;
 170 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
 171         &prot_fault_translation, 0, "Select signal to deliver on protection fault");
 172 static int uprintf_signal;
 173 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW,
 174     &uprintf_signal, 0,
 175     "Print debugging information on trap signal to ctty");
 176
 177 /*
 178  * Exception, fault, and trap interface to the FreeBSD kernel.
 179  * This common code is called from assembly language IDT gate entry
 180  * routines that prepare a suitable stack frame, and restore this
 181  * frame after the exception has been processed.
 182  */
 183
 184 void
 185 trap(struct trapframe *frame)
 186 {
 187 #ifdef KDTRACE_HOOKS
 188         struct reg regs;
 189 #endif
 190         struct thread *td = curthread;
 191         struct proc *p = td->td_proc;
 192         int i = 0, ucode = 0, code;
 193         u_int type;
 194         register_t addr = 0;
 195         vm_offset_t eva;
 196         ksiginfo_t ksi;
 197 #ifdef POWERFAIL_NMI
 198         static int lastalert = 0;
 199 #endif
 200
 201         PCPU_INC(cnt.v_trap);
 202         type = frame->tf_trapno;
 203
 204 #ifdef SMP
 205         /* Handler for NMI IPIs used for stopping CPUs. */
 206         if (type == T_NMI) {
 207                  if (ipi_nmi_handler() == 0)
 208                            goto out;
 209         }
 210 #endif /* SMP */
 211
 212 #ifdef KDB
 213         if (kdb_active) {
 214                 kdb_reenter();
 215                 goto out;
 216         }
 217 #endif
 218
 219         if (type == T_RESERVED) {
 220                 trap_fatal(frame, 0);
 221                 goto out;
 222         }
 223
 224         if (type == T_NMI) {
 225 #ifdef HWPMC_HOOKS
 226                 /*
 227                  * CPU PMCs interrupt using an NMI so we check for that first.
 228                  * If the HWPMC module is active, 'pmc_hook' will point to
 229                  * the function to be called.  A non-zero return value from the
 230                  * hook means that the NMI was consumed by it and that we can
 231                  * return immediately.
 232                  */
 233                 if (pmc_intr != NULL &&
 234                     (*pmc_intr)(PCPU_GET(cpuid), frame) != 0)
 235                         goto out;
 236 #endif
 237
 238 #ifdef STACK
 239                 if (stack_nmi_handler(frame) != 0)
 240                         goto out;
 241 #endif
 242         }
 243
 244         if (type == T_MCHK) {
 245                 mca_intr();
 246                 goto out;
 247         }
 248
 249 #ifdef KDTRACE_HOOKS
 250         /*
 251          * A trap can occur while DTrace executes a probe. Before
 252          * executing the probe, DTrace blocks re-scheduling and sets
 253          * a flag in its per-cpu flags to indicate that it doesn't
 254          * want to fault. On returning from the probe, the no-fault
 255          * flag is cleared and finally re-scheduling is enabled.
 256          */
 257         if ((type == T_PROTFLT || type == T_PAGEFLT) &&
 258             dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
 259                 goto out;
 260 #endif
 261
 262         if ((frame->tf_eflags & PSL_I) == 0) {
 263                 /*
 264                  * Buggy application or kernel code has disabled
 265                  * interrupts and then trapped.  Enabling interrupts
 266                  * now is wrong, but it is better than running with
 267                  * interrupts disabled until they are accidentally
 268                  * enabled later.
 269                  */
 270                 if (ISPL(frame->tf_cs) == SEL_UPL || (frame->tf_eflags & PSL_VM))
 271                         uprintf(
 272                             "pid %ld (%s): trap %d with interrupts disabled\n",
 273                             (long)curproc->p_pid, curthread->td_name, type);
 274                 else if (type != T_NMI && type != T_BPTFLT &&
 275                     type != T_TRCTRAP &&
 276                     frame->tf_eip != (int)cpu_switch_load_gs) {
 277                         /*
 278                          * XXX not quite right, since this may be for a
 279                          * multiple fault in user mode.
 280                          */
 281                         printf("kernel trap %d with interrupts disabled\n",
 282                             type);
 283                         /*
 284                          * Page faults need interrupts disabled until later,
 285                          * and we shouldn't enable interrupts while holding
 286                          * a spin lock.
 287                          */
 288                         if (type != T_PAGEFLT &&
 289                             td->td_md.md_spinlock_count == 0)
 290                                 enable_intr();
 291                 }
 292         }
 293         eva = 0;
 294         code = frame->tf_err;
 295         if (type == T_PAGEFLT) {
 296                 /*
 297                  * For some Cyrix CPUs, %cr2 is clobbered by
 298                  * interrupts.  This problem is worked around by using
 299                  * an interrupt gate for the pagefault handler.  We
 300                  * are finally ready to read %cr2 and conditionally
 301                  * reenable interrupts.  If we hold a spin lock, then
 302                  * we must not reenable interrupts.  This might be a
 303                  * spurious page fault.
 304                  */
 305                 eva = rcr2();
 306                 if (td->td_md.md_spinlock_count == 0)
 307                         enable_intr();
 308         }
 309
 310         if ((ISPL(frame->tf_cs) == SEL_UPL) ||
 311             ((frame->tf_eflags & PSL_VM) &&
 312                 !(curpcb->pcb_flags & PCB_VM86CALL))) {
 313                 /* user trap */
 314
 315                 td->td_pticks = 0;
 316                 td->td_frame = frame;
 317                 addr = frame->tf_eip;
 318                 if (td->td_cowgen != p->p_cowgen)
 319                         thread_cow_update(td);
 320
 321                 switch (type) {
 322                 case T_PRIVINFLT:       /* privileged instruction fault */
 323                         i = SIGILL;
 324                         ucode = ILL_PRVOPC;
 325                         break;
 326
 327                 case T_BPTFLT:          /* bpt instruction fault */
 328                 case T_TRCTRAP:         /* trace trap */
 329                         enable_intr();
 330 #ifdef KDTRACE_HOOKS
 331                         if (type == T_BPTFLT) {
 332                                 fill_frame_regs(frame, &regs);
 333                                 if (dtrace_pid_probe_ptr != NULL &&
 334                                     dtrace_pid_probe_ptr(&regs) == 0)
 335                                         goto out;
 336                         }
 337 #endif
 338                         frame->tf_eflags &= ~PSL_T;
 339                         i = SIGTRAP;
 340                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 341                         break;
 342
 343                 case T_ARITHTRAP:       /* arithmetic trap */
 344 #ifdef DEV_NPX
 345                         ucode = npxtrap_x87();
 346                         if (ucode == -1)
 347                                 goto userout;
 348 #else
 349                         ucode = 0;
 350 #endif
 351                         i = SIGFPE;
 352                         break;
 353
 354                         /*
 355                          * The following two traps can happen in
 356                          * vm86 mode, and, if so, we want to handle
 357                          * them specially.
 358                          */
 359                 case T_PROTFLT:         /* general protection fault */
 360                 case T_STKFLT:          /* stack fault */
 361                         if (frame->tf_eflags & PSL_VM) {
 362                                 i = vm86_emulate((struct vm86frame *)frame);
 363                                 if (i == 0)
 364                                         goto user;
 365                                 break;
 366                         }
 367                         i = SIGBUS;
 368                         ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
 369                         break;
 370                 case T_SEGNPFLT:        /* segment not present fault */
 371                         i = SIGBUS;
 372                         ucode = BUS_ADRERR;
 373                         break;
 374                 case T_TSSFLT:          /* invalid TSS fault */
 375                         i = SIGBUS;
 376                         ucode = BUS_OBJERR;
 377                         break;
 378                 case T_ALIGNFLT:
 379                         i = SIGBUS;
 380                         ucode = BUS_ADRALN;
 381                         break;
 382                 case T_DOUBLEFLT:       /* double fault */
 383                 default:
 384                         i = SIGBUS;
 385                         ucode = BUS_OBJERR;
 386                         break;
 387
 388                 case T_PAGEFLT:         /* page fault */
 389
 390                         i = trap_pfault(frame, TRUE, eva);
 391 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 392                         if (i == -2) {
 393                                 /*
 394                                  * The f00f hack workaround has triggered, so
 395                                  * treat the fault as an illegal instruction
 396                                  * (T_PRIVINFLT) instead of a page fault.
 397                                  */
 398                                 type = frame->tf_trapno = T_PRIVINFLT;
 399
 400                                 /* Proceed as in that case. */
 401                                 ucode = ILL_PRVOPC;
 402                                 i = SIGILL;
 403                                 break;
 404                         }
 405 #endif
 406                         if (i == -1)
 407                                 goto userout;
 408                         if (i == 0)
 409                                 goto user;
 410
 411                         if (i == SIGSEGV)
 412                                 ucode = SEGV_MAPERR;
 413                         else {
 414                                 if (prot_fault_translation == 0) {
 415                                         /*
 416                                          * Autodetect.
 417                                          * This check also covers the images
 418                                          * without the ABI-tag ELF note.
 419                                          */
 420                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 421                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 422                                                 i = SIGSEGV;
 423                                                 ucode = SEGV_ACCERR;
 424                                         } else {
 425                                                 i = SIGBUS;
 426                                                 ucode = BUS_PAGE_FAULT;
 427                                         }
 428                                 } else if (prot_fault_translation == 1) {
 429                                         /*
 430                                          * Always compat mode.
 431                                          */
 432                                         i = SIGBUS;
 433                                         ucode = BUS_PAGE_FAULT;
 434                                 } else {
 435                                         /*
 436                                          * Always SIGSEGV mode.
 437                                          */
 438                                         i = SIGSEGV;
 439                                         ucode = SEGV_ACCERR;
 440                                 }
 441                         }
 442                         addr = eva;
 443                         break;
 444
 445                 case T_DIVIDE:          /* integer divide fault */
 446                         ucode = FPE_INTDIV;
 447                         i = SIGFPE;
 448                         break;
 449
 450 #ifdef DEV_ISA
 451                 case T_NMI:
 452 #ifdef POWERFAIL_NMI
 453 #ifndef TIMER_FREQ
 454 #  define TIMER_FREQ 1193182
 455 #endif
 456                         if (time_second - lastalert > 10) {
 457                                 log(LOG_WARNING, "NMI: power fail\n");
 458                                 sysbeep(880, hz);
 459                                 lastalert = time_second;
 460                         }
 461                         goto userout;
 462 #else /* !POWERFAIL_NMI */
 463                         /* machine/parity/power fail/"kitchen sink" faults */
 464                         if (isa_nmi(code) == 0) {
 465 #ifdef KDB
 466                                 /*
 467                                  * NMI can be hooked up to a pushbutton
 468                                  * for debugging.
 469                                  */
 470                                 if (kdb_on_nmi) {
 471                                         printf ("NMI ... going to debugger\n");
 472                                         kdb_trap(type, 0, frame);
 473                                 }
 474 #endif /* KDB */
 475                                 goto userout;
 476                         } else if (panic_on_nmi)
 477                                 panic("NMI indicates hardware failure");
 478                         break;
 479 #endif /* POWERFAIL_NMI */
 480 #endif /* DEV_ISA */
 481
 482                 case T_OFLOW:           /* integer overflow fault */
 483                         ucode = FPE_INTOVF;
 484                         i = SIGFPE;
 485                         break;
 486
 487                 case T_BOUND:           /* bounds check fault */
 488                         ucode = FPE_FLTSUB;
 489                         i = SIGFPE;
 490                         break;
 491
 492                 case T_DNA:
 493 #ifdef DEV_NPX
 494                         KASSERT(PCB_USER_FPU(td->td_pcb),
 495                             ("kernel FPU ctx has leaked"));
 496                         /* transparent fault (due to context switch "late") */
 497                         if (npxdna())
 498                                 goto userout;
 499 #endif
 500                         uprintf("pid %d killed due to lack of floating point\n",
 501                                 p->p_pid);
 502                         i = SIGKILL;
 503                         ucode = 0;
 504                         break;
 505
 506                 case T_FPOPFLT:         /* FPU operand fetch fault */
 507                         ucode = ILL_COPROC;
 508                         i = SIGILL;
 509                         break;
 510
 511                 case T_XMMFLT:          /* SIMD floating-point exception */
 512 #if defined(DEV_NPX) && !defined(CPU_DISABLE_SSE) && defined(I686_CPU)
 513                         ucode = npxtrap_sse();
 514                         if (ucode == -1)
 515                                 goto userout;
 516 #else
 517                         ucode = 0;
 518 #endif
 519                         i = SIGFPE;
 520                         break;
 521 #ifdef KDTRACE_HOOKS
 522                 case T_DTRACE_RET:
 523                         enable_intr();
 524                         fill_frame_regs(frame, &regs);
 525                         if (dtrace_return_probe_ptr != NULL &&
 526                             dtrace_return_probe_ptr(&regs) == 0)
 527                                 goto out;
 528                         break;
 529 #endif
 530                 }
 531         } else {
 532                 /* kernel trap */
 533
 534                 KASSERT(cold || td->td_ucred != NULL,
 535                     ("kernel trap doesn't have ucred"));
 536                 switch (type) {
 537                 case T_PAGEFLT:                 /* page fault */
 538                         (void) trap_pfault(frame, FALSE, eva);
 539                         goto out;
 540
 541                 case T_DNA:
 542 #ifdef DEV_NPX
 543                         if (PCB_USER_FPU(td->td_pcb))
 544                                 panic("Unregistered use of FPU in kernel");
 545                         if (npxdna())
 546                                 goto out;
 547 #endif
 548                         break;
 549
 550                 case T_ARITHTRAP:       /* arithmetic trap */
 551                 case T_XMMFLT:          /* SIMD floating-point exception */
 552                 case T_FPOPFLT:         /* FPU operand fetch fault */
 553                         /*
 554                          * XXXKIB for now disable any FPU traps in kernel
 555                          * handler registration seems to be overkill
 556                          */
 557                         trap_fatal(frame, 0);
 558                         goto out;
 559
 560                         /*
 561                          * The following two traps can happen in
 562                          * vm86 mode, and, if so, we want to handle
 563                          * them specially.
 564                          */
 565                 case T_PROTFLT:         /* general protection fault */
 566                 case T_STKFLT:          /* stack fault */
 567                         if (frame->tf_eflags & PSL_VM) {
 568                                 i = vm86_emulate((struct vm86frame *)frame);
 569                                 if (i != 0)
 570                                         /*
 571                                          * returns to original process
 572                                          */
 573                                         vm86_trap((struct vm86frame *)frame);
 574                                 goto out;
 575                         }
 576                         if (type == T_STKFLT)
 577                                 break;
 578
 579                         /* FALL THROUGH */
 580
 581                 case T_SEGNPFLT:        /* segment not present fault */
 582                         if (curpcb->pcb_flags & PCB_VM86CALL)
 583                                 break;
 584
 585                         /*
 586                          * Invalid %fs's and %gs's can be created using
 587                          * procfs or PT_SETREGS or by invalidating the
 588                          * underlying LDT entry.  This causes a fault
 589                          * in kernel mode when the kernel attempts to
 590                          * switch contexts.  Lose the bad context
 591                          * (XXX) so that we can continue, and generate
 592                          * a signal.
 593                          */
 594                         if (frame->tf_eip == (int)cpu_switch_load_gs) {
 595                                 curpcb->pcb_gs = 0;
 596 #if 0
 597                                 PROC_LOCK(p);
 598                                 kern_psignal(p, SIGBUS);
 599                                 PROC_UNLOCK(p);
 600 #endif
 601                                 goto out;
 602                         }
 603
 604                         if (td->td_intr_nesting_level != 0)
 605                                 break;
 606
 607                         /*
 608                          * Invalid segment selectors and out of bounds
 609                          * %eip's and %esp's can be set up in user mode.
 610                          * This causes a fault in kernel mode when the
 611                          * kernel tries to return to user mode.  We want
 612                          * to get this fault so that we can fix the
 613                          * problem here and not have to check all the
 614                          * selectors and pointers when the user changes
 615                          * them.
 616                          */
 617                         if (frame->tf_eip == (int)doreti_iret) {
 618                                 frame->tf_eip = (int)doreti_iret_fault;
 619                                 goto out;
 620                         }
 621                         if (frame->tf_eip == (int)doreti_popl_ds) {
 622                                 frame->tf_eip = (int)doreti_popl_ds_fault;
 623                                 goto out;
 624                         }
 625                         if (frame->tf_eip == (int)doreti_popl_es) {
 626                                 frame->tf_eip = (int)doreti_popl_es_fault;
 627                                 goto out;
 628                         }
 629                         if (frame->tf_eip == (int)doreti_popl_fs) {
 630                                 frame->tf_eip = (int)doreti_popl_fs_fault;
 631                                 goto out;
 632                         }
 633                         if (curpcb->pcb_onfault != NULL) {
 634                                 frame->tf_eip =
 635                                     (int)curpcb->pcb_onfault;
 636                                 goto out;
 637                         }
 638                         break;
 639
 640                 case T_TSSFLT:
 641                         /*
 642                          * PSL_NT can be set in user mode and isn't cleared
 643                          * automatically when the kernel is entered.  This
 644                          * causes a TSS fault when the kernel attempts to
 645                          * `iret' because the TSS link is uninitialized.  We
 646                          * want to get this fault so that we can fix the
 647                          * problem here and not every time the kernel is
 648                          * entered.
 649                          */
 650                         if (frame->tf_eflags & PSL_NT) {
 651                                 frame->tf_eflags &= ~PSL_NT;
 652                                 goto out;
 653                         }
 654                         break;
 655
 656                 case T_TRCTRAP:  /* trace trap */
 657                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall)) {
 658                                 /*
 659                                  * We've just entered system mode via the
 660                                  * syscall lcall.  Continue single stepping
 661                                  * silently until the syscall handler has
 662                                  * saved the flags.
 663                                  */
 664                                 goto out;
 665                         }
 666                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall) + 1) {
 667                                 /*
 668                                  * The syscall handler has now saved the
 669                                  * flags.  Stop single stepping it.
 670                                  */
 671                                 frame->tf_eflags &= ~PSL_T;
 672                                 goto out;
 673                         }
 674                         /*
 675                          * Ignore debug register trace traps due to
 676                          * accesses in the user's address space, which
 677                          * can happen under several conditions such as
 678                          * if a user sets a watchpoint on a buffer and
 679                          * then passes that buffer to a system call.
 680                          * We still want to get TRCTRAPS for addresses
 681                          * in kernel space because that is useful when
 682                          * debugging the kernel.
 683                          */
 684                         if (user_dbreg_trap() &&
 685                            !(curpcb->pcb_flags & PCB_VM86CALL)) {
 686                                 /*
 687                                  * Reset breakpoint bits because the
 688                                  * processor doesn't
 689                                  */
 690                                 load_dr6(rdr6() & 0xfffffff0);
 691                                 goto out;
 692                         }
 693                         /*
 694                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 695                          */
 696                 case T_BPTFLT:
 697                         /*
 698                          * If KDB is enabled, let it handle the debugger trap.
 699                          * Otherwise, debugger traps "can't happen".
 700                          */
 701 #ifdef KDB
 702                         if (kdb_trap(type, 0, frame))
 703                                 goto out;
 704 #endif
 705                         break;
 706
 707 #ifdef DEV_ISA
 708                 case T_NMI:
 709 #ifdef POWERFAIL_NMI
 710                         if (time_second - lastalert > 10) {
 711                                 log(LOG_WARNING, "NMI: power fail\n");
 712                                 sysbeep(880, hz);
 713                                 lastalert = time_second;
 714                         }
 715                         goto out;
 716 #else /* !POWERFAIL_NMI */
 717                         /* machine/parity/power fail/"kitchen sink" faults */
 718                         if (isa_nmi(code) == 0) {
 719 #ifdef KDB
 720                                 /*
 721                                  * NMI can be hooked up to a pushbutton
 722                                  * for debugging.
 723                                  */
 724                                 if (kdb_on_nmi) {
 725                                         printf ("NMI ... going to debugger\n");
 726                                         kdb_trap(type, 0, frame);
 727                                 }
 728 #endif /* KDB */
 729                                 goto out;
 730                         } else if (panic_on_nmi == 0)
 731                                 goto out;
 732                         /* FALLTHROUGH */
 733 #endif /* POWERFAIL_NMI */
 734 #endif /* DEV_ISA */
 735                 }
 736
 737                 trap_fatal(frame, eva);
 738                 goto out;
 739         }
 740
 741         /* Translate fault for emulators (e.g. Linux) */
 742         if (*p->p_sysent->sv_transtrap)
 743                 i = (*p->p_sysent->sv_transtrap)(i, type);
 744
 745         ksiginfo_init_trap(&ksi);
 746         ksi.ksi_signo = i;
 747         ksi.ksi_code = ucode;
 748         ksi.ksi_addr = (void *)addr;
 749         ksi.ksi_trapno = type;
 750         if (uprintf_signal) {
 751                 uprintf("pid %d comm %s: signal %d err %x code %d type %d "
 752                     "addr 0x%x esp 0x%08x eip 0x%08x "
 753                     "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
 754                     p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr,
 755                     frame->tf_esp, frame->tf_eip,
 756                     fubyte((void *)(frame->tf_eip + 0)),
 757                     fubyte((void *)(frame->tf_eip + 1)),
 758                     fubyte((void *)(frame->tf_eip + 2)),
 759                     fubyte((void *)(frame->tf_eip + 3)),
 760                     fubyte((void *)(frame->tf_eip + 4)),
 761                     fubyte((void *)(frame->tf_eip + 5)),
 762                     fubyte((void *)(frame->tf_eip + 6)),
 763                     fubyte((void *)(frame->tf_eip + 7)));
 764         }
 765         KASSERT((read_eflags() & PSL_I) != 0, ("interrupts disabled"));
 766         trapsignal(td, &ksi);
 767
 768 #ifdef DEBUG
 769         if (type <= MAX_TRAP_MSG) {
 770                 uprintf("fatal process exception: %s",
 771                         trap_msg[type]);
 772                 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
 773                         uprintf(", fault VA = 0x%lx", (u_long)eva);
 774                 uprintf("\n");
 775         }
 776 #endif
 777
 778 user:
 779         userret(td, frame);
 780         KASSERT(PCB_USER_FPU(td->td_pcb),
 781             ("Return from trap with kernel FPU ctx leaked"));
 782 userout:
 783 out:
 784         return;
 785 }
 786
 787 static int
 788 trap_pfault(frame, usermode, eva)
 789         struct trapframe *frame;
 790         int usermode;
 791         vm_offset_t eva;
 792 {
 793         vm_offset_t va;
 794         vm_map_t map;
 795         int rv = 0;
 796         vm_prot_t ftype;
 797         struct thread *td = curthread;
 798         struct proc *p = td->td_proc;
 799
 800         if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
 801                 /*
 802                  * Due to both processor errata and lazy TLB invalidation when
 803                  * access restrictions are removed from virtual pages, memory
 804                  * accesses that are allowed by the physical mapping layer may
 805                  * nonetheless cause one spurious page fault per virtual page.
 806                  * When the thread is executing a "no faulting" section that
 807                  * is bracketed by vm_fault_{disable,enable}_pagefaults(),
 808                  * every page fault is treated as a spurious page fault,
 809                  * unless it accesses the same virtual address as the most
 810                  * recent page fault within the same "no faulting" section.
 811                  */
 812                 if (td->td_md.md_spurflt_addr != eva ||
 813                     (td->td_pflags & TDP_RESETSPUR) != 0) {
 814                         /*
 815                          * Do nothing to the TLB.  A stale TLB entry is
 816                          * flushed automatically by a page fault.
 817                          */
 818                         td->td_md.md_spurflt_addr = eva;
 819                         td->td_pflags &= ~TDP_RESETSPUR;
 820                         return (0);
 821                 }
 822         } else {
 823                 /*
 824                  * If we get a page fault while in a critical section, then
 825                  * it is most likely a fatal kernel page fault.  The kernel
 826                  * is already going to panic trying to get a sleep lock to
 827                  * do the VM lookup, so just consider it a fatal trap so the
 828                  * kernel can print out a useful trap message and even get
 829                  * to the debugger.
 830                  *
 831                  * If we get a page fault while holding a non-sleepable
 832                  * lock, then it is most likely a fatal kernel page fault.
 833                  * If WITNESS is enabled, then it's going to whine about
 834                  * bogus LORs with various VM locks, so just skip to the
 835                  * fatal trap handling directly.
 836                  */
 837                 if (td->td_critnest != 0 ||
 838                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 839                     "Kernel page fault") != 0) {
 840                         trap_fatal(frame, eva);
 841                         return (-1);
 842                 }
 843         }
 844         va = trunc_page(eva);
 845         if (va >= KERNBASE) {
 846                 /*
 847                  * Don't allow user-mode faults in kernel address space.
 848                  * An exception:  if the faulting address is the invalid
 849                  * instruction entry in the IDT, then the Intel Pentium
 850                  * F00F bug workaround was triggered, and we need to
 851                  * treat it is as an illegal instruction, and not a page
 852                  * fault.
 853                  */
 854 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 855                 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
 856                         return (-2);
 857 #endif
 858                 if (usermode)
 859                         goto nogo;
 860
 861                 map = kernel_map;
 862         } else {
 863                 map = &p->p_vmspace->vm_map;
 864
 865                 /*
 866                  * When accessing a user-space address, kernel must be
 867                  * ready to accept the page fault, and provide a
 868                  * handling routine.  Since accessing the address
 869                  * without the handler is a bug, do not try to handle
 870                  * it normally, and panic immediately.
 871                  */
 872                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 873                     curpcb->pcb_onfault == NULL)) {
 874                         trap_fatal(frame, eva);
 875                         return (-1);
 876                 }
 877         }
 878
 879         /*
 880          * PGEX_I is defined only if the execute disable bit capability is
 881          * supported and enabled.
 882          */
 883         if (frame->tf_err & PGEX_W)
 884                 ftype = VM_PROT_WRITE;
 885 #if defined(PAE) || defined(PAE_TABLES)
 886         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 887                 ftype = VM_PROT_EXECUTE;
 888 #endif
 889         else
 890                 ftype = VM_PROT_READ;
 891
 892         /* Fault in the page. */
 893         rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 894         if (rv == KERN_SUCCESS) {
 895 #ifdef HWPMC_HOOKS
 896                 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
 897                         PMC_SOFT_CALL_TF( , , page_fault, all, frame);
 898                         if (ftype == VM_PROT_READ)
 899                                 PMC_SOFT_CALL_TF( , , page_fault, read,
 900                                     frame);
 901                         else
 902                                 PMC_SOFT_CALL_TF( , , page_fault, write,
 903                                     frame);
 904                 }
 905 #endif
 906                 return (0);
 907         }
 908 nogo:
 909         if (!usermode) {
 910                 if (td->td_intr_nesting_level == 0 &&
 911                     curpcb->pcb_onfault != NULL) {
 912                         frame->tf_eip = (int)curpcb->pcb_onfault;
 913                         return (0);
 914                 }
 915                 trap_fatal(frame, eva);
 916                 return (-1);
 917         }
 918         return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 919 }
 920
 921 static void
 922 trap_fatal(frame, eva)
 923         struct trapframe *frame;
 924         vm_offset_t eva;
 925 {
 926         int code, ss, esp;
 927         u_int type;
 928         struct soft_segment_descriptor softseg;
 929         char *msg;
 930
 931         code = frame->tf_err;
 932         type = frame->tf_trapno;
 933         sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
 934
 935         if (type <= MAX_TRAP_MSG)
 936                 msg = trap_msg[type];
 937         else
 938                 msg = "UNKNOWN";
 939         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 940             frame->tf_eflags & PSL_VM ? "vm86" :
 941             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
 942 #ifdef SMP
 943         /* two separate prints in case of a trap on an unmapped page */
 944         printf("cpuid = %d; ", PCPU_GET(cpuid));
 945         printf("apic id = %02x\n", PCPU_GET(apic_id));
 946 #endif
 947         if (type == T_PAGEFLT) {
 948                 printf("fault virtual address   = 0x%x\n", eva);
 949                 printf("fault code              = %s %s, %s\n",
 950                         code & PGEX_U ? "user" : "supervisor",
 951                         code & PGEX_W ? "write" : "read",
 952                         code & PGEX_P ? "protection violation" : "page not present");
 953         }
 954         printf("instruction pointer     = 0x%x:0x%x\n",
 955                frame->tf_cs & 0xffff, frame->tf_eip);
 956         if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
 957                 ss = frame->tf_ss & 0xffff;
 958                 esp = frame->tf_esp;
 959         } else {
 960                 ss = GSEL(GDATA_SEL, SEL_KPL);
 961                 esp = (int)&frame->tf_esp;
 962         }
 963         printf("stack pointer           = 0x%x:0x%x\n", ss, esp);
 964         printf("frame pointer           = 0x%x:0x%x\n", ss, frame->tf_ebp);
 965         printf("code segment            = base 0x%x, limit 0x%x, type 0x%x\n",
 966                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 967         printf("                        = DPL %d, pres %d, def32 %d, gran %d\n",
 968                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
 969                softseg.ssd_gran);
 970         printf("processor eflags        = ");
 971         if (frame->tf_eflags & PSL_T)
 972                 printf("trace trap, ");
 973         if (frame->tf_eflags & PSL_I)
 974                 printf("interrupt enabled, ");
 975         if (frame->tf_eflags & PSL_NT)
 976                 printf("nested task, ");
 977         if (frame->tf_eflags & PSL_RF)
 978                 printf("resume, ");
 979         if (frame->tf_eflags & PSL_VM)
 980                 printf("vm86, ");
 981         printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
 982         printf("current process         = %d (%s)\n",
 983             curproc->p_pid, curthread->td_name);
 984
 985 #ifdef KDB
 986         if (debugger_on_panic || kdb_active) {
 987                 frame->tf_err = eva;    /* smuggle fault address to ddb */
 988                 if (kdb_trap(type, 0, frame)) {
 989                         frame->tf_err = code;   /* restore error code */
 990                         return;
 991                 }
 992                 frame->tf_err = code;           /* restore error code */
 993         }
 994 #endif
 995         printf("trap number             = %d\n", type);
 996         if (type <= MAX_TRAP_MSG)
 997                 panic("%s", trap_msg[type]);
 998         else
 999                 panic("unknown/reserved trap");
1000 }
1001
1002 /*
1003  * Double fault handler. Called when a fault occurs while writing
1004  * a frame for a trap/exception onto the stack. This usually occurs
1005  * when the stack overflows (such is the case with infinite recursion,
1006  * for example).
1007  *
1008  * XXX Note that the current PTD gets replaced by IdlePTD when the
1009  * task switch occurs. This means that the stack that was active at
1010  * the time of the double fault is not available at <kstack> unless
1011  * the machine was idle when the double fault occurred. The downside
1012  * of this is that "trace <ebp>" in ddb won't work.
1013  */
1014 void
1015 dblfault_handler()
1016 {
1017 #ifdef KDTRACE_HOOKS
1018         if (dtrace_doubletrap_func != NULL)
1019                 (*dtrace_doubletrap_func)();
1020 #endif
1021         printf("\nFatal double fault:\n");
1022         printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip));
1023         printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp));
1024         printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp));
1025 #ifdef SMP
1026         /* two separate prints in case of a trap on an unmapped page */
1027         printf("cpuid = %d; ", PCPU_GET(cpuid));
1028         printf("apic id = %02x\n", PCPU_GET(apic_id));
1029 #endif
1030         panic("double fault");
1031 }
1032
1033 int
1034 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
1035 {
1036         struct proc *p;
1037         struct trapframe *frame;
1038         caddr_t params;
1039         long tmp;
1040         int error;
1041
1042         p = td->td_proc;
1043         frame = td->td_frame;
1044
1045         params = (caddr_t)frame->tf_esp + sizeof(int);
1046         sa->code = frame->tf_eax;
1047
1048         /*
1049          * Need to check if this is a 32 bit or 64 bit syscall.
1050          */
1051         if (sa->code == SYS_syscall) {
1052                 /*
1053                  * Code is first argument, followed by actual args.
1054                  */
1055                 error = fueword(params, &tmp);
1056                 if (error == -1)
1057                         return (EFAULT);
1058                 sa->code = tmp;
1059                 params += sizeof(int);
1060         } else if (sa->code == SYS___syscall) {
1061                 /*
1062                  * Like syscall, but code is a quad, so as to maintain
1063                  * quad alignment for the rest of the arguments.
1064                  */
1065                 error = fueword(params, &tmp);
1066                 if (error == -1)
1067                         return (EFAULT);
1068                 sa->code = tmp;
1069                 params += sizeof(quad_t);
1070         }
1071
1072         if (p->p_sysent->sv_mask)
1073                 sa->code &= p->p_sysent->sv_mask;
1074         if (sa->code >= p->p_sysent->sv_size)
1075                 sa->callp = &p->p_sysent->sv_table[0];
1076         else
1077                 sa->callp = &p->p_sysent->sv_table[sa->code];
1078         sa->narg = sa->callp->sy_narg;
1079
1080         if (params != NULL && sa->narg != 0)
1081                 error = copyin(params, (caddr_t)sa->args,
1082                     (u_int)(sa->narg * sizeof(int)));
1083         else
1084                 error = 0;
1085
1086         if (error == 0) {
1087                 td->td_retval[0] = 0;
1088                 td->td_retval[1] = frame->tf_edx;
1089         }
1090
1091         return (error);
1092 }
1093
1094 #include "../../kern/subr_syscall.c"
1095
1096 /*
1097  * syscall - system call request C handler.  A system call is
1098  * essentially treated as a trap by reusing the frame layout.
1099  */
1100 void
1101 syscall(struct trapframe *frame)
1102 {
1103         struct thread *td;
1104         struct syscall_args sa;
1105         register_t orig_tf_eflags;
1106         int error;
1107         ksiginfo_t ksi;
1108
1109 #ifdef DIAGNOSTIC
1110         if (ISPL(frame->tf_cs) != SEL_UPL) {
1111                 panic("syscall");
1112                 /* NOT REACHED */
1113         }
1114 #endif
1115         orig_tf_eflags = frame->tf_eflags;
1116
1117         td = curthread;
1118         td->td_frame = frame;
1119
1120         error = syscallenter(td, &sa);
1121
1122         /*
1123          * Traced syscall.
1124          */
1125         if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1126                 frame->tf_eflags &= ~PSL_T;
1127                 ksiginfo_init_trap(&ksi);
1128                 ksi.ksi_signo = SIGTRAP;
1129                 ksi.ksi_code = TRAP_TRACE;
1130                 ksi.ksi_addr = (void *)frame->tf_eip;
1131                 trapsignal(td, &ksi);
1132         }
1133
1134         KASSERT(PCB_USER_FPU(td->td_pcb),
1135             ("System call %s returning with kernel FPU ctx leaked",
1136              syscallname(td->td_proc, sa.code)));
1137         KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td),
1138             ("System call %s returning with mangled pcb_save",
1139              syscallname(td->td_proc, sa.code)));
1140
1141         syscallret(td, error, &sa);
1142 }