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_kdtrace.h"
  53 #include "opt_npx.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/tss.h>
  99 #include <machine/vm86.h>
 100
 101 #ifdef POWERFAIL_NMI
 102 #include <sys/syslog.h>
 103 #include <machine/clock.h>
 104 #endif
 105
 106 #ifdef KDTRACE_HOOKS
 107 #include <sys/dtrace_bsd.h>
 108 #endif
 109
 110 extern void trap(struct trapframe *frame);
 111 extern void syscall(struct trapframe *frame);
 112
 113 static int trap_pfault(struct trapframe *, int, vm_offset_t);
 114 static void trap_fatal(struct trapframe *, vm_offset_t);
 115 void dblfault_handler(void);
 116
 117 extern inthand_t IDTVEC(lcall_syscall);
 118
 119 #define MAX_TRAP_MSG            32
 120 static char *trap_msg[] = {
 121         "",                                     /*  0 unused */
 122         "privileged instruction fault",         /*  1 T_PRIVINFLT */
 123         "",                                     /*  2 unused */
 124         "breakpoint instruction fault",         /*  3 T_BPTFLT */
 125         "",                                     /*  4 unused */
 126         "",                                     /*  5 unused */
 127         "arithmetic trap",                      /*  6 T_ARITHTRAP */
 128         "",                                     /*  7 unused */
 129         "",                                     /*  8 unused */
 130         "general protection fault",             /*  9 T_PROTFLT */
 131         "trace trap",                           /* 10 T_TRCTRAP */
 132         "",                                     /* 11 unused */
 133         "page fault",                           /* 12 T_PAGEFLT */
 134         "",                                     /* 13 unused */
 135         "alignment fault",                      /* 14 T_ALIGNFLT */
 136         "",                                     /* 15 unused */
 137         "",                                     /* 16 unused */
 138         "",                                     /* 17 unused */
 139         "integer divide fault",                 /* 18 T_DIVIDE */
 140         "non-maskable interrupt trap",          /* 19 T_NMI */
 141         "overflow trap",                        /* 20 T_OFLOW */
 142         "FPU bounds check fault",               /* 21 T_BOUND */
 143         "FPU device not available",             /* 22 T_DNA */
 144         "double fault",                         /* 23 T_DOUBLEFLT */
 145         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
 146         "invalid TSS fault",                    /* 25 T_TSSFLT */
 147         "segment not present fault",            /* 26 T_SEGNPFLT */
 148         "stack fault",                          /* 27 T_STKFLT */
 149         "machine check trap",                   /* 28 T_MCHK */
 150         "SIMD floating-point exception",        /* 29 T_XMMFLT */
 151         "reserved (unknown) fault",             /* 30 T_RESERVED */
 152         "",                                     /* 31 unused (reserved) */
 153         "DTrace pid return trap",               /* 32 T_DTRACE_RET */
 154 };
 155
 156 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 157 int has_f00f_bug = 0;           /* Initialized so that it can be patched. */
 158 #endif
 159
 160 #ifdef KDB
 161 static int kdb_on_nmi = 1;
 162 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
 163         &kdb_on_nmi, 0, "Go to KDB on NMI");
 164 TUNABLE_INT("machdep.kdb_on_nmi", &kdb_on_nmi);
 165 #endif
 166 static int panic_on_nmi = 1;
 167 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
 168         &panic_on_nmi, 0, "Panic on NMI");
 169 TUNABLE_INT("machdep.panic_on_nmi", &panic_on_nmi);
 170 static int prot_fault_translation = 0;
 171 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
 172         &prot_fault_translation, 0, "Select signal to deliver on protection fault");
 173 static int uprintf_signal;
 174 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW,
 175     &uprintf_signal, 0,
 176     "Print debugging information on trap signal to ctty");
 177
 178 /*
 179  * Exception, fault, and trap interface to the FreeBSD kernel.
 180  * This common code is called from assembly language IDT gate entry
 181  * routines that prepare a suitable stack frame, and restore this
 182  * frame after the exception has been processed.
 183  */
 184
 185 void
 186 trap(struct trapframe *frame)
 187 {
 188 #ifdef KDTRACE_HOOKS
 189         struct reg regs;
 190 #endif
 191         struct thread *td = curthread;
 192         struct proc *p = td->td_proc;
 193         int i = 0, ucode = 0, code;
 194         u_int type;
 195         register_t addr = 0;
 196         vm_offset_t eva;
 197         ksiginfo_t ksi;
 198 #ifdef POWERFAIL_NMI
 199         static int lastalert = 0;
 200 #endif
 201
 202         PCPU_INC(cnt.v_trap);
 203         type = frame->tf_trapno;
 204
 205 #ifdef SMP
 206         /* Handler for NMI IPIs used for stopping CPUs. */
 207         if (type == T_NMI) {
 208                  if (ipi_nmi_handler() == 0)
 209                            goto out;
 210         }
 211 #endif /* SMP */
 212
 213 #ifdef KDB
 214         if (kdb_active) {
 215                 kdb_reenter();
 216                 goto out;
 217         }
 218 #endif
 219
 220         if (type == T_RESERVED) {
 221                 trap_fatal(frame, 0);
 222                 goto out;
 223         }
 224
 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 return value of '1' from the
 230          * hook means that the NMI was handled by it and that we can
 231          * return immediately.
 232          */
 233         if (type == T_NMI && pmc_intr &&
 234             (*pmc_intr)(PCPU_GET(cpuid), frame))
 235             goto out;
 236 #endif
 237
 238         if (type == T_MCHK) {
 239                 mca_intr();
 240                 goto out;
 241         }
 242
 243 #ifdef KDTRACE_HOOKS
 244         /*
 245          * A trap can occur while DTrace executes a probe. Before
 246          * executing the probe, DTrace blocks re-scheduling and sets
 247          * a flag in its per-cpu flags to indicate that it doesn't
 248          * want to fault. On returning from the probe, the no-fault
 249          * flag is cleared and finally re-scheduling is enabled.
 250          */
 251         if ((type == T_PROTFLT || type == T_PAGEFLT) &&
 252             dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
 253                 goto out;
 254 #endif
 255
 256         if ((frame->tf_eflags & PSL_I) == 0) {
 257                 /*
 258                  * Buggy application or kernel code has disabled
 259                  * interrupts and then trapped.  Enabling interrupts
 260                  * now is wrong, but it is better than running with
 261                  * interrupts disabled until they are accidentally
 262                  * enabled later.
 263                  */
 264                 if (ISPL(frame->tf_cs) == SEL_UPL || (frame->tf_eflags & PSL_VM))
 265                         uprintf(
 266                             "pid %ld (%s): trap %d with interrupts disabled\n",
 267                             (long)curproc->p_pid, curthread->td_name, type);
 268                 else if (type != T_NMI && type != T_BPTFLT &&
 269                     type != T_TRCTRAP &&
 270                     frame->tf_eip != (int)cpu_switch_load_gs) {
 271                         /*
 272                          * XXX not quite right, since this may be for a
 273                          * multiple fault in user mode.
 274                          */
 275                         printf("kernel trap %d with interrupts disabled\n",
 276                             type);
 277                         /*
 278                          * Page faults need interrupts disabled until later,
 279                          * and we shouldn't enable interrupts while holding
 280                          * a spin lock.
 281                          */
 282                         if (type != T_PAGEFLT &&
 283                             td->td_md.md_spinlock_count == 0)
 284                                 enable_intr();
 285                 }
 286         }
 287         eva = 0;
 288         code = frame->tf_err;
 289         if (type == T_PAGEFLT) {
 290                 /*
 291                  * For some Cyrix CPUs, %cr2 is clobbered by
 292                  * interrupts.  This problem is worked around by using
 293                  * an interrupt gate for the pagefault handler.  We
 294                  * are finally ready to read %cr2 and conditionally
 295                  * reenable interrupts.  If we hold a spin lock, then
 296                  * we must not reenable interrupts.  This might be a
 297                  * spurious page fault.
 298                  */
 299                 eva = rcr2();
 300                 if (td->td_md.md_spinlock_count == 0)
 301                         enable_intr();
 302         }
 303
 304         if ((ISPL(frame->tf_cs) == SEL_UPL) ||
 305             ((frame->tf_eflags & PSL_VM) &&
 306                 !(curpcb->pcb_flags & PCB_VM86CALL))) {
 307                 /* user trap */
 308
 309                 td->td_pticks = 0;
 310                 td->td_frame = frame;
 311                 addr = frame->tf_eip;
 312                 if (td->td_ucred != p->p_ucred)
 313                         cred_update_thread(td);
 314
 315                 switch (type) {
 316                 case T_PRIVINFLT:       /* privileged instruction fault */
 317                         i = SIGILL;
 318                         ucode = ILL_PRVOPC;
 319                         break;
 320
 321                 case T_BPTFLT:          /* bpt instruction fault */
 322                 case T_TRCTRAP:         /* trace trap */
 323                         enable_intr();
 324 #ifdef KDTRACE_HOOKS
 325                         if (type == T_BPTFLT) {
 326                                 fill_frame_regs(frame, &regs);
 327                                 if (dtrace_pid_probe_ptr != NULL &&
 328                                     dtrace_pid_probe_ptr(&regs) == 0)
 329                                         goto out;
 330                         }
 331 #endif
 332                         frame->tf_eflags &= ~PSL_T;
 333                         i = SIGTRAP;
 334                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 335                         break;
 336
 337                 case T_ARITHTRAP:       /* arithmetic trap */
 338 #ifdef DEV_NPX
 339                         ucode = npxtrap_x87();
 340                         if (ucode == -1)
 341                                 goto userout;
 342 #else
 343                         ucode = 0;
 344 #endif
 345                         i = SIGFPE;
 346                         break;
 347
 348                         /*
 349                          * The following two traps can happen in
 350                          * vm86 mode, and, if so, we want to handle
 351                          * them specially.
 352                          */
 353                 case T_PROTFLT:         /* general protection fault */
 354                 case T_STKFLT:          /* stack fault */
 355                         if (frame->tf_eflags & PSL_VM) {
 356                                 i = vm86_emulate((struct vm86frame *)frame);
 357                                 if (i == 0)
 358                                         goto user;
 359                                 break;
 360                         }
 361                         i = SIGBUS;
 362                         ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
 363                         break;
 364                 case T_SEGNPFLT:        /* segment not present fault */
 365                         i = SIGBUS;
 366                         ucode = BUS_ADRERR;
 367                         break;
 368                 case T_TSSFLT:          /* invalid TSS fault */
 369                         i = SIGBUS;
 370                         ucode = BUS_OBJERR;
 371                         break;
 372                 case T_ALIGNFLT:
 373                         i = SIGBUS;
 374                         ucode = BUS_ADRALN;
 375                         break;
 376                 case T_DOUBLEFLT:       /* double fault */
 377                 default:
 378                         i = SIGBUS;
 379                         ucode = BUS_OBJERR;
 380                         break;
 381
 382                 case T_PAGEFLT:         /* page fault */
 383
 384                         i = trap_pfault(frame, TRUE, eva);
 385 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 386                         if (i == -2) {
 387                                 /*
 388                                  * The f00f hack workaround has triggered, so
 389                                  * treat the fault as an illegal instruction
 390                                  * (T_PRIVINFLT) instead of a page fault.
 391                                  */
 392                                 type = frame->tf_trapno = T_PRIVINFLT;
 393
 394                                 /* Proceed as in that case. */
 395                                 ucode = ILL_PRVOPC;
 396                                 i = SIGILL;
 397                                 break;
 398                         }
 399 #endif
 400                         if (i == -1)
 401                                 goto userout;
 402                         if (i == 0)
 403                                 goto user;
 404
 405                         if (i == SIGSEGV)
 406                                 ucode = SEGV_MAPERR;
 407                         else {
 408                                 if (prot_fault_translation == 0) {
 409                                         /*
 410                                          * Autodetect.
 411                                          * This check also covers the images
 412                                          * without the ABI-tag ELF note.
 413                                          */
 414                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 415                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 416                                                 i = SIGSEGV;
 417                                                 ucode = SEGV_ACCERR;
 418                                         } else {
 419                                                 i = SIGBUS;
 420                                                 ucode = BUS_PAGE_FAULT;
 421                                         }
 422                                 } else if (prot_fault_translation == 1) {
 423                                         /*
 424                                          * Always compat mode.
 425                                          */
 426                                         i = SIGBUS;
 427                                         ucode = BUS_PAGE_FAULT;
 428                                 } else {
 429                                         /*
 430                                          * Always SIGSEGV mode.
 431                                          */
 432                                         i = SIGSEGV;
 433                                         ucode = SEGV_ACCERR;
 434                                 }
 435                         }
 436                         addr = eva;
 437                         break;
 438
 439                 case T_DIVIDE:          /* integer divide fault */
 440                         ucode = FPE_INTDIV;
 441                         i = SIGFPE;
 442                         break;
 443
 444 #ifdef DEV_ISA
 445                 case T_NMI:
 446 #ifdef POWERFAIL_NMI
 447 #ifndef TIMER_FREQ
 448 #  define TIMER_FREQ 1193182
 449 #endif
 450                         if (time_second - lastalert > 10) {
 451                                 log(LOG_WARNING, "NMI: power fail\n");
 452                                 sysbeep(880, hz);
 453                                 lastalert = time_second;
 454                         }
 455                         goto userout;
 456 #else /* !POWERFAIL_NMI */
 457                         /* machine/parity/power fail/"kitchen sink" faults */
 458                         if (isa_nmi(code) == 0) {
 459 #ifdef KDB
 460                                 /*
 461                                  * NMI can be hooked up to a pushbutton
 462                                  * for debugging.
 463                                  */
 464                                 if (kdb_on_nmi) {
 465                                         printf ("NMI ... going to debugger\n");
 466                                         kdb_trap(type, 0, frame);
 467                                 }
 468 #endif /* KDB */
 469                                 goto userout;
 470                         } else if (panic_on_nmi)
 471                                 panic("NMI indicates hardware failure");
 472                         break;
 473 #endif /* POWERFAIL_NMI */
 474 #endif /* DEV_ISA */
 475
 476                 case T_OFLOW:           /* integer overflow fault */
 477                         ucode = FPE_INTOVF;
 478                         i = SIGFPE;
 479                         break;
 480
 481                 case T_BOUND:           /* bounds check fault */
 482                         ucode = FPE_FLTSUB;
 483                         i = SIGFPE;
 484                         break;
 485
 486                 case T_DNA:
 487 #ifdef DEV_NPX
 488                         KASSERT(PCB_USER_FPU(td->td_pcb),
 489                             ("kernel FPU ctx has leaked"));
 490                         /* transparent fault (due to context switch "late") */
 491                         if (npxdna())
 492                                 goto userout;
 493 #endif
 494                         uprintf("pid %d killed due to lack of floating point\n",
 495                                 p->p_pid);
 496                         i = SIGKILL;
 497                         ucode = 0;
 498                         break;
 499
 500                 case T_FPOPFLT:         /* FPU operand fetch fault */
 501                         ucode = ILL_COPROC;
 502                         i = SIGILL;
 503                         break;
 504
 505                 case T_XMMFLT:          /* SIMD floating-point exception */
 506 #if defined(DEV_NPX) && !defined(CPU_DISABLE_SSE) && defined(I686_CPU)
 507                         ucode = npxtrap_sse();
 508                         if (ucode == -1)
 509                                 goto userout;
 510 #else
 511                         ucode = 0;
 512 #endif
 513                         i = SIGFPE;
 514                         break;
 515 #ifdef KDTRACE_HOOKS
 516                 case T_DTRACE_RET:
 517                         enable_intr();
 518                         fill_frame_regs(frame, &regs);
 519                         if (dtrace_return_probe_ptr != NULL &&
 520                             dtrace_return_probe_ptr(&regs) == 0)
 521                                 goto out;
 522                         break;
 523 #endif
 524                 }
 525         } else {
 526                 /* kernel trap */
 527
 528                 KASSERT(cold || td->td_ucred != NULL,
 529                     ("kernel trap doesn't have ucred"));
 530                 switch (type) {
 531                 case T_PAGEFLT:                 /* page fault */
 532                         (void) trap_pfault(frame, FALSE, eva);
 533                         goto out;
 534
 535                 case T_DNA:
 536 #ifdef DEV_NPX
 537                         KASSERT(!PCB_USER_FPU(td->td_pcb),
 538                             ("Unregistered use of FPU in kernel"));
 539                         if (npxdna())
 540                                 goto out;
 541 #endif
 542                         break;
 543
 544                 case T_ARITHTRAP:       /* arithmetic trap */
 545                 case T_XMMFLT:          /* SIMD floating-point exception */
 546                 case T_FPOPFLT:         /* FPU operand fetch fault */
 547                         /*
 548                          * XXXKIB for now disable any FPU traps in kernel
 549                          * handler registration seems to be overkill
 550                          */
 551                         trap_fatal(frame, 0);
 552                         goto out;
 553
 554                         /*
 555                          * The following two traps can happen in
 556                          * vm86 mode, and, if so, we want to handle
 557                          * them specially.
 558                          */
 559                 case T_PROTFLT:         /* general protection fault */
 560                 case T_STKFLT:          /* stack fault */
 561                         if (frame->tf_eflags & PSL_VM) {
 562                                 i = vm86_emulate((struct vm86frame *)frame);
 563                                 if (i != 0)
 564                                         /*
 565                                          * returns to original process
 566                                          */
 567                                         vm86_trap((struct vm86frame *)frame);
 568                                 goto out;
 569                         }
 570                         if (type == T_STKFLT)
 571                                 break;
 572
 573                         /* FALL THROUGH */
 574
 575                 case T_SEGNPFLT:        /* segment not present fault */
 576                         if (curpcb->pcb_flags & PCB_VM86CALL)
 577                                 break;
 578
 579                         /*
 580                          * Invalid %fs's and %gs's can be created using
 581                          * procfs or PT_SETREGS or by invalidating the
 582                          * underlying LDT entry.  This causes a fault
 583                          * in kernel mode when the kernel attempts to
 584                          * switch contexts.  Lose the bad context
 585                          * (XXX) so that we can continue, and generate
 586                          * a signal.
 587                          */
 588                         if (frame->tf_eip == (int)cpu_switch_load_gs) {
 589                                 curpcb->pcb_gs = 0;
 590 #if 0
 591                                 PROC_LOCK(p);
 592                                 kern_psignal(p, SIGBUS);
 593                                 PROC_UNLOCK(p);
 594 #endif
 595                                 goto out;
 596                         }
 597
 598                         if (td->td_intr_nesting_level != 0)
 599                                 break;
 600
 601                         /*
 602                          * Invalid segment selectors and out of bounds
 603                          * %eip's and %esp's can be set up in user mode.
 604                          * This causes a fault in kernel mode when the
 605                          * kernel tries to return to user mode.  We want
 606                          * to get this fault so that we can fix the
 607                          * problem here and not have to check all the
 608                          * selectors and pointers when the user changes
 609                          * them.
 610                          */
 611                         if (frame->tf_eip == (int)doreti_iret) {
 612                                 frame->tf_eip = (int)doreti_iret_fault;
 613                                 goto out;
 614                         }
 615                         if (frame->tf_eip == (int)doreti_popl_ds) {
 616                                 frame->tf_eip = (int)doreti_popl_ds_fault;
 617                                 goto out;
 618                         }
 619                         if (frame->tf_eip == (int)doreti_popl_es) {
 620                                 frame->tf_eip = (int)doreti_popl_es_fault;
 621                                 goto out;
 622                         }
 623                         if (frame->tf_eip == (int)doreti_popl_fs) {
 624                                 frame->tf_eip = (int)doreti_popl_fs_fault;
 625                                 goto out;
 626                         }
 627                         if (curpcb->pcb_onfault != NULL) {
 628                                 frame->tf_eip =
 629                                     (int)curpcb->pcb_onfault;
 630                                 goto out;
 631                         }
 632                         break;
 633
 634                 case T_TSSFLT:
 635                         /*
 636                          * PSL_NT can be set in user mode and isn't cleared
 637                          * automatically when the kernel is entered.  This
 638                          * causes a TSS fault when the kernel attempts to
 639                          * `iret' because the TSS link is uninitialized.  We
 640                          * want to get this fault so that we can fix the
 641                          * problem here and not every time the kernel is
 642                          * entered.
 643                          */
 644                         if (frame->tf_eflags & PSL_NT) {
 645                                 frame->tf_eflags &= ~PSL_NT;
 646                                 goto out;
 647                         }
 648                         break;
 649
 650                 case T_TRCTRAP:  /* trace trap */
 651                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall)) {
 652                                 /*
 653                                  * We've just entered system mode via the
 654                                  * syscall lcall.  Continue single stepping
 655                                  * silently until the syscall handler has
 656                                  * saved the flags.
 657                                  */
 658                                 goto out;
 659                         }
 660                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall) + 1) {
 661                                 /*
 662                                  * The syscall handler has now saved the
 663                                  * flags.  Stop single stepping it.
 664                                  */
 665                                 frame->tf_eflags &= ~PSL_T;
 666                                 goto out;
 667                         }
 668                         /*
 669                          * Ignore debug register trace traps due to
 670                          * accesses in the user's address space, which
 671                          * can happen under several conditions such as
 672                          * if a user sets a watchpoint on a buffer and
 673                          * then passes that buffer to a system call.
 674                          * We still want to get TRCTRAPS for addresses
 675                          * in kernel space because that is useful when
 676                          * debugging the kernel.
 677                          */
 678                         if (user_dbreg_trap() &&
 679                            !(curpcb->pcb_flags & PCB_VM86CALL)) {
 680                                 /*
 681                                  * Reset breakpoint bits because the
 682                                  * processor doesn't
 683                                  */
 684                                 load_dr6(rdr6() & 0xfffffff0);
 685                                 goto out;
 686                         }
 687                         /*
 688                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 689                          */
 690                 case T_BPTFLT:
 691                         /*
 692                          * If KDB is enabled, let it handle the debugger trap.
 693                          * Otherwise, debugger traps "can't happen".
 694                          */
 695 #ifdef KDB
 696                         if (kdb_trap(type, 0, frame))
 697                                 goto out;
 698 #endif
 699                         break;
 700
 701 #ifdef DEV_ISA
 702                 case T_NMI:
 703 #ifdef POWERFAIL_NMI
 704                         if (time_second - lastalert > 10) {
 705                                 log(LOG_WARNING, "NMI: power fail\n");
 706                                 sysbeep(880, hz);
 707                                 lastalert = time_second;
 708                         }
 709                         goto out;
 710 #else /* !POWERFAIL_NMI */
 711                         /* machine/parity/power fail/"kitchen sink" faults */
 712                         if (isa_nmi(code) == 0) {
 713 #ifdef KDB
 714                                 /*
 715                                  * NMI can be hooked up to a pushbutton
 716                                  * for debugging.
 717                                  */
 718                                 if (kdb_on_nmi) {
 719                                         printf ("NMI ... going to debugger\n");
 720                                         kdb_trap(type, 0, frame);
 721                                 }
 722 #endif /* KDB */
 723                                 goto out;
 724                         } else if (panic_on_nmi == 0)
 725                                 goto out;
 726                         /* FALLTHROUGH */
 727 #endif /* POWERFAIL_NMI */
 728 #endif /* DEV_ISA */
 729                 }
 730
 731                 trap_fatal(frame, eva);
 732                 goto out;
 733         }
 734
 735         /* Translate fault for emulators (e.g. Linux) */
 736         if (*p->p_sysent->sv_transtrap)
 737                 i = (*p->p_sysent->sv_transtrap)(i, type);
 738
 739         ksiginfo_init_trap(&ksi);
 740         ksi.ksi_signo = i;
 741         ksi.ksi_code = ucode;
 742         ksi.ksi_addr = (void *)addr;
 743         ksi.ksi_trapno = type;
 744         if (uprintf_signal) {
 745                 uprintf("pid %d comm %s: signal %d err %x code %d type %d "
 746                     "addr 0x%x esp 0x%08x eip 0x%08x "
 747                     "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
 748                     p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr,
 749                     frame->tf_esp, frame->tf_eip,
 750                     fubyte((void *)(frame->tf_eip + 0)),
 751                     fubyte((void *)(frame->tf_eip + 1)),
 752                     fubyte((void *)(frame->tf_eip + 2)),
 753                     fubyte((void *)(frame->tf_eip + 3)),
 754                     fubyte((void *)(frame->tf_eip + 4)),
 755                     fubyte((void *)(frame->tf_eip + 5)),
 756                     fubyte((void *)(frame->tf_eip + 6)),
 757                     fubyte((void *)(frame->tf_eip + 7)));
 758         }
 759         KASSERT((read_eflags() & PSL_I) != 0, ("interrupts disabled"));
 760         trapsignal(td, &ksi);
 761
 762 #ifdef DEBUG
 763         if (type <= MAX_TRAP_MSG) {
 764                 uprintf("fatal process exception: %s",
 765                         trap_msg[type]);
 766                 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
 767                         uprintf(", fault VA = 0x%lx", (u_long)eva);
 768                 uprintf("\n");
 769         }
 770 #endif
 771
 772 user:
 773         userret(td, frame);
 774         KASSERT(PCB_USER_FPU(td->td_pcb),
 775             ("Return from trap with kernel FPU ctx leaked"));
 776 userout:
 777 out:
 778         return;
 779 }
 780
 781 static int
 782 trap_pfault(frame, usermode, eva)
 783         struct trapframe *frame;
 784         int usermode;
 785         vm_offset_t eva;
 786 {
 787         vm_offset_t va;
 788         struct vmspace *vm;
 789         vm_map_t map;
 790         int rv = 0;
 791         vm_prot_t ftype;
 792         struct thread *td = curthread;
 793         struct proc *p = td->td_proc;
 794
 795         if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
 796                 /*
 797                  * Due to both processor errata and lazy TLB invalidation when
 798                  * access restrictions are removed from virtual pages, memory
 799                  * accesses that are allowed by the physical mapping layer may
 800                  * nonetheless cause one spurious page fault per virtual page.
 801                  * When the thread is executing a "no faulting" section that
 802                  * is bracketed by vm_fault_{disable,enable}_pagefaults(),
 803                  * every page fault is treated as a spurious page fault,
 804                  * unless it accesses the same virtual address as the most
 805                  * recent page fault within the same "no faulting" section.
 806                  */
 807                 if (td->td_md.md_spurflt_addr != eva ||
 808                     (td->td_pflags & TDP_RESETSPUR) != 0) {
 809                         /*
 810                          * Do nothing to the TLB.  A stale TLB entry is
 811                          * flushed automatically by a page fault.
 812                          */
 813                         td->td_md.md_spurflt_addr = eva;
 814                         td->td_pflags &= ~TDP_RESETSPUR;
 815                         return (0);
 816                 }
 817         } else {
 818                 /*
 819                  * If we get a page fault while in a critical section, then
 820                  * it is most likely a fatal kernel page fault.  The kernel
 821                  * is already going to panic trying to get a sleep lock to
 822                  * do the VM lookup, so just consider it a fatal trap so the
 823                  * kernel can print out a useful trap message and even get
 824                  * to the debugger.
 825                  *
 826                  * If we get a page fault while holding a non-sleepable
 827                  * lock, then it is most likely a fatal kernel page fault.
 828                  * If WITNESS is enabled, then it's going to whine about
 829                  * bogus LORs with various VM locks, so just skip to the
 830                  * fatal trap handling directly.
 831                  */
 832                 if (td->td_critnest != 0 ||
 833                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 834                     "Kernel page fault") != 0) {
 835                         trap_fatal(frame, eva);
 836                         return (-1);
 837                 }
 838         }
 839         va = trunc_page(eva);
 840         if (va >= KERNBASE) {
 841                 /*
 842                  * Don't allow user-mode faults in kernel address space.
 843                  * An exception:  if the faulting address is the invalid
 844                  * instruction entry in the IDT, then the Intel Pentium
 845                  * F00F bug workaround was triggered, and we need to
 846                  * treat it is as an illegal instruction, and not a page
 847                  * fault.
 848                  */
 849 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 850                 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
 851                         return (-2);
 852 #endif
 853                 if (usermode)
 854                         goto nogo;
 855
 856                 map = kernel_map;
 857         } else {
 858                 /*
 859                  * This is a fault on non-kernel virtual memory.  If either
 860                  * p or p->p_vmspace is NULL, then the fault is fatal.
 861                  */
 862                 if (p == NULL || (vm = p->p_vmspace) == NULL)
 863                         goto nogo;
 864
 865                 map = &vm->vm_map;
 866
 867                 /*
 868                  * When accessing a user-space address, kernel must be
 869                  * ready to accept the page fault, and provide a
 870                  * handling routine.  Since accessing the address
 871                  * without the handler is a bug, do not try to handle
 872                  * it normally, and panic immediately.
 873                  */
 874                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 875                     curpcb->pcb_onfault == NULL)) {
 876                         trap_fatal(frame, eva);
 877                         return (-1);
 878                 }
 879         }
 880
 881         /*
 882          * PGEX_I is defined only if the execute disable bit capability is
 883          * supported and enabled.
 884          */
 885         if (frame->tf_err & PGEX_W)
 886                 ftype = VM_PROT_WRITE;
 887 #if defined(PAE) || defined(PAE_TABLES)
 888         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 889                 ftype = VM_PROT_EXECUTE;
 890 #endif
 891         else
 892                 ftype = VM_PROT_READ;
 893
 894         if (map != kernel_map) {
 895                 /*
 896                  * Keep swapout from messing with us during this
 897                  *      critical time.
 898                  */
 899                 PROC_LOCK(p);
 900                 ++p->p_lock;
 901                 PROC_UNLOCK(p);
 902
 903                 /* Fault in the user page: */
 904                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 905
 906                 PROC_LOCK(p);
 907                 --p->p_lock;
 908                 PROC_UNLOCK(p);
 909         } else {
 910                 /*
 911                  * Don't have to worry about process locking or stacks in the
 912                  * kernel.
 913                  */
 914                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 915         }
 916         if (rv == KERN_SUCCESS) {
 917 #ifdef HWPMC_HOOKS
 918                 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
 919                         PMC_SOFT_CALL_TF( , , page_fault, all, frame);
 920                         if (ftype == VM_PROT_READ)
 921                                 PMC_SOFT_CALL_TF( , , page_fault, read,
 922                                     frame);
 923                         else
 924                                 PMC_SOFT_CALL_TF( , , page_fault, write,
 925                                     frame);
 926                 }
 927 #endif
 928                 return (0);
 929         }
 930 nogo:
 931         if (!usermode) {
 932                 if (td->td_intr_nesting_level == 0 &&
 933                     curpcb->pcb_onfault != NULL) {
 934                         frame->tf_eip = (int)curpcb->pcb_onfault;
 935                         return (0);
 936                 }
 937                 trap_fatal(frame, eva);
 938                 return (-1);
 939         }
 940         return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 941 }
 942
 943 static void
 944 trap_fatal(frame, eva)
 945         struct trapframe *frame;
 946         vm_offset_t eva;
 947 {
 948         int code, ss, esp;
 949         u_int type;
 950         struct soft_segment_descriptor softseg;
 951         char *msg;
 952
 953         code = frame->tf_err;
 954         type = frame->tf_trapno;
 955         sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
 956
 957         if (type <= MAX_TRAP_MSG)
 958                 msg = trap_msg[type];
 959         else
 960                 msg = "UNKNOWN";
 961         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 962             frame->tf_eflags & PSL_VM ? "vm86" :
 963             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
 964 #ifdef SMP
 965         /* two separate prints in case of a trap on an unmapped page */
 966         printf("cpuid = %d; ", PCPU_GET(cpuid));
 967         printf("apic id = %02x\n", PCPU_GET(apic_id));
 968 #endif
 969         if (type == T_PAGEFLT) {
 970                 printf("fault virtual address   = 0x%x\n", eva);
 971                 printf("fault code              = %s %s, %s\n",
 972                         code & PGEX_U ? "user" : "supervisor",
 973                         code & PGEX_W ? "write" : "read",
 974                         code & PGEX_P ? "protection violation" : "page not present");
 975         }
 976         printf("instruction pointer     = 0x%x:0x%x\n",
 977                frame->tf_cs & 0xffff, frame->tf_eip);
 978         if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
 979                 ss = frame->tf_ss & 0xffff;
 980                 esp = frame->tf_esp;
 981         } else {
 982                 ss = GSEL(GDATA_SEL, SEL_KPL);
 983                 esp = (int)&frame->tf_esp;
 984         }
 985         printf("stack pointer           = 0x%x:0x%x\n", ss, esp);
 986         printf("frame pointer           = 0x%x:0x%x\n", ss, frame->tf_ebp);
 987         printf("code segment            = base 0x%x, limit 0x%x, type 0x%x\n",
 988                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 989         printf("                        = DPL %d, pres %d, def32 %d, gran %d\n",
 990                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
 991                softseg.ssd_gran);
 992         printf("processor eflags        = ");
 993         if (frame->tf_eflags & PSL_T)
 994                 printf("trace trap, ");
 995         if (frame->tf_eflags & PSL_I)
 996                 printf("interrupt enabled, ");
 997         if (frame->tf_eflags & PSL_NT)
 998                 printf("nested task, ");
 999         if (frame->tf_eflags & PSL_RF)
1000                 printf("resume, ");
1001         if (frame->tf_eflags & PSL_VM)
1002                 printf("vm86, ");
1003         printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1004         printf("current process         = %d (%s)\n",
1005             curproc->p_pid, curthread->td_name);
1006
1007 #ifdef KDB
1008         if (debugger_on_panic || kdb_active) {
1009                 frame->tf_err = eva;    /* smuggle fault address to ddb */
1010                 if (kdb_trap(type, 0, frame)) {
1011                         frame->tf_err = code;   /* restore error code */
1012                         return;
1013                 }
1014                 frame->tf_err = code;           /* restore error code */
1015         }
1016 #endif
1017         printf("trap number             = %d\n", type);
1018         if (type <= MAX_TRAP_MSG)
1019                 panic("%s", trap_msg[type]);
1020         else
1021                 panic("unknown/reserved trap");
1022 }
1023
1024 /*
1025  * Double fault handler. Called when a fault occurs while writing
1026  * a frame for a trap/exception onto the stack. This usually occurs
1027  * when the stack overflows (such is the case with infinite recursion,
1028  * for example).
1029  *
1030  * XXX Note that the current PTD gets replaced by IdlePTD when the
1031  * task switch occurs. This means that the stack that was active at
1032  * the time of the double fault is not available at <kstack> unless
1033  * the machine was idle when the double fault occurred. The downside
1034  * of this is that "trace <ebp>" in ddb won't work.
1035  */
1036 void
1037 dblfault_handler()
1038 {
1039 #ifdef KDTRACE_HOOKS
1040         if (dtrace_doubletrap_func != NULL)
1041                 (*dtrace_doubletrap_func)();
1042 #endif
1043         printf("\nFatal double fault:\n");
1044         printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip));
1045         printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp));
1046         printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp));
1047 #ifdef SMP
1048         /* two separate prints in case of a trap on an unmapped page */
1049         printf("cpuid = %d; ", PCPU_GET(cpuid));
1050         printf("apic id = %02x\n", PCPU_GET(apic_id));
1051 #endif
1052         panic("double fault");
1053 }
1054
1055 int
1056 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
1057 {
1058         struct proc *p;
1059         struct trapframe *frame;
1060         caddr_t params;
1061         long tmp;
1062         int error;
1063
1064         p = td->td_proc;
1065         frame = td->td_frame;
1066
1067         params = (caddr_t)frame->tf_esp + sizeof(int);
1068         sa->code = frame->tf_eax;
1069
1070         /*
1071          * Need to check if this is a 32 bit or 64 bit syscall.
1072          */
1073         if (sa->code == SYS_syscall) {
1074                 /*
1075                  * Code is first argument, followed by actual args.
1076                  */
1077                 error = fueword(params, &tmp);
1078                 if (error == -1)
1079                         return (EFAULT);
1080                 sa->code = tmp;
1081                 params += sizeof(int);
1082         } else if (sa->code == SYS___syscall) {
1083                 /*
1084                  * Like syscall, but code is a quad, so as to maintain
1085                  * quad alignment for the rest of the arguments.
1086                  */
1087                 error = fueword(params, &tmp);
1088                 if (error == -1)
1089                         return (EFAULT);
1090                 sa->code = tmp;
1091                 params += sizeof(quad_t);
1092         }
1093
1094         if (p->p_sysent->sv_mask)
1095                 sa->code &= p->p_sysent->sv_mask;
1096         if (sa->code >= p->p_sysent->sv_size)
1097                 sa->callp = &p->p_sysent->sv_table[0];
1098         else
1099                 sa->callp = &p->p_sysent->sv_table[sa->code];
1100         sa->narg = sa->callp->sy_narg;
1101
1102         if (params != NULL && sa->narg != 0)
1103                 error = copyin(params, (caddr_t)sa->args,
1104                     (u_int)(sa->narg * sizeof(int)));
1105         else
1106                 error = 0;
1107
1108         if (error == 0) {
1109                 td->td_retval[0] = 0;
1110                 td->td_retval[1] = frame->tf_edx;
1111         }
1112
1113         return (error);
1114 }
1115
1116 #include "../../kern/subr_syscall.c"
1117
1118 /*
1119  * syscall - system call request C handler.  A system call is
1120  * essentially treated as a trap by reusing the frame layout.
1121  */
1122 void
1123 syscall(struct trapframe *frame)
1124 {
1125         struct thread *td;
1126         struct syscall_args sa;
1127         register_t orig_tf_eflags;
1128         int error;
1129         ksiginfo_t ksi;
1130
1131 #ifdef DIAGNOSTIC
1132         if (ISPL(frame->tf_cs) != SEL_UPL) {
1133                 panic("syscall");
1134                 /* NOT REACHED */
1135         }
1136 #endif
1137         orig_tf_eflags = frame->tf_eflags;
1138
1139         td = curthread;
1140         td->td_frame = frame;
1141
1142         error = syscallenter(td, &sa);
1143
1144         /*
1145          * Traced syscall.
1146          */
1147         if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1148                 frame->tf_eflags &= ~PSL_T;
1149                 ksiginfo_init_trap(&ksi);
1150                 ksi.ksi_signo = SIGTRAP;
1151                 ksi.ksi_code = TRAP_TRACE;
1152                 ksi.ksi_addr = (void *)frame->tf_eip;
1153                 trapsignal(td, &ksi);
1154         }
1155
1156         KASSERT(PCB_USER_FPU(td->td_pcb),
1157             ("System call %s returning with kernel FPU ctx leaked",
1158              syscallname(td->td_proc, sa.code)));
1159         KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td),
1160             ("System call %s returning with mangled pcb_save",
1161              syscallname(td->td_proc, sa.code)));
1162
1163         syscallret(td, error, &sa);
1164 }