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_ktrace.h"
  54 #include "opt_npx.h"
  55 #include "opt_trap.h"
  56
  57 #include <sys/param.h>
  58 #include <sys/bus.h>
  59 #include <sys/systm.h>
  60 #include <sys/proc.h>
  61 #include <sys/pioctl.h>
  62 #include <sys/ptrace.h>
  63 #include <sys/kdb.h>
  64 #include <sys/kernel.h>
  65 #include <sys/ktr.h>
  66 #include <sys/lock.h>
  67 #include <sys/mutex.h>
  68 #include <sys/resourcevar.h>
  69 #include <sys/signalvar.h>
  70 #include <sys/syscall.h>
  71 #include <sys/sysctl.h>
  72 #include <sys/sysent.h>
  73 #include <sys/uio.h>
  74 #include <sys/vmmeter.h>
  75 #ifdef KTRACE
  76 #include <sys/ktrace.h>
  77 #endif
  78 #ifdef HWPMC_HOOKS
  79 #include <sys/pmckern.h>
  80 #endif
  81 #include <security/audit/audit.h>
  82
  83 #include <vm/vm.h>
  84 #include <vm/vm_param.h>
  85 #include <vm/pmap.h>
  86 #include <vm/vm_kern.h>
  87 #include <vm/vm_map.h>
  88 #include <vm/vm_page.h>
  89 #include <vm/vm_extern.h>
  90
  91 #include <machine/cpu.h>
  92 #include <machine/intr_machdep.h>
  93 #include <machine/mca.h>
  94 #include <machine/md_var.h>
  95 #include <machine/pcb.h>
  96 #ifdef SMP
  97 #include <machine/smp.h>
  98 #endif
  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
 110 /*
 111  * This is a hook which is initialised by the dtrace module
 112  * to handle traps which might occur during DTrace probe
 113  * execution.
 114  */
 115 dtrace_trap_func_t      dtrace_trap_func;
 116
 117 dtrace_doubletrap_func_t        dtrace_doubletrap_func;
 118
 119 /*
 120  * This is a hook which is initialised by the systrace module
 121  * when it is loaded. This keeps the DTrace syscall provider
 122  * implementation opaque.
 123  */
 124 systrace_probe_func_t   systrace_probe_func;
 125 #endif
 126
 127 extern void trap(struct trapframe *frame);
 128 extern void syscall(struct trapframe *frame);
 129
 130 static int trap_pfault(struct trapframe *, int, vm_offset_t);
 131 static void trap_fatal(struct trapframe *, vm_offset_t);
 132 void dblfault_handler(void);
 133
 134 extern inthand_t IDTVEC(lcall_syscall);
 135
 136 #define MAX_TRAP_MSG            30
 137 static char *trap_msg[] = {
 138         "",                                     /*  0 unused */
 139         "privileged instruction fault",         /*  1 T_PRIVINFLT */
 140         "",                                     /*  2 unused */
 141         "breakpoint instruction fault",         /*  3 T_BPTFLT */
 142         "",                                     /*  4 unused */
 143         "",                                     /*  5 unused */
 144         "arithmetic trap",                      /*  6 T_ARITHTRAP */
 145         "",                                     /*  7 unused */
 146         "",                                     /*  8 unused */
 147         "general protection fault",             /*  9 T_PROTFLT */
 148         "trace trap",                           /* 10 T_TRCTRAP */
 149         "",                                     /* 11 unused */
 150         "page fault",                           /* 12 T_PAGEFLT */
 151         "",                                     /* 13 unused */
 152         "alignment fault",                      /* 14 T_ALIGNFLT */
 153         "",                                     /* 15 unused */
 154         "",                                     /* 16 unused */
 155         "",                                     /* 17 unused */
 156         "integer divide fault",                 /* 18 T_DIVIDE */
 157         "non-maskable interrupt trap",          /* 19 T_NMI */
 158         "overflow trap",                        /* 20 T_OFLOW */
 159         "FPU bounds check fault",               /* 21 T_BOUND */
 160         "FPU device not available",             /* 22 T_DNA */
 161         "double fault",                         /* 23 T_DOUBLEFLT */
 162         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
 163         "invalid TSS fault",                    /* 25 T_TSSFLT */
 164         "segment not present fault",            /* 26 T_SEGNPFLT */
 165         "stack fault",                          /* 27 T_STKFLT */
 166         "machine check trap",                   /* 28 T_MCHK */
 167         "SIMD floating-point exception",        /* 29 T_XMMFLT */
 168         "reserved (unknown) fault",             /* 30 T_RESERVED */
 169 };
 170
 171 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 172 extern int has_f00f_bug;
 173 #endif
 174
 175 #ifdef KDB
 176 static int kdb_on_nmi = 1;
 177 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
 178         &kdb_on_nmi, 0, "Go to KDB on NMI");
 179 #endif
 180 static int panic_on_nmi = 1;
 181 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
 182         &panic_on_nmi, 0, "Panic on NMI");
 183 static int prot_fault_translation = 0;
 184 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
 185         &prot_fault_translation, 0, "Select signal to deliver on protection fault");
 186
 187 /*
 188  * Exception, fault, and trap interface to the FreeBSD kernel.
 189  * This common code is called from assembly language IDT gate entry
 190  * routines that prepare a suitable stack frame, and restore this
 191  * frame after the exception has been processed.
 192  */
 193
 194 void
 195 trap(struct trapframe *frame)
 196 {
 197         struct thread *td = curthread;
 198         struct proc *p = td->td_proc;
 199         int i = 0, ucode = 0, code;
 200         u_int type;
 201         register_t addr = 0;
 202         vm_offset_t eva;
 203         ksiginfo_t ksi;
 204 #ifdef POWERFAIL_NMI
 205         static int lastalert = 0;
 206 #endif
 207
 208         PCPU_INC(cnt.v_trap);
 209         type = frame->tf_trapno;
 210
 211 #ifdef SMP
 212         /* Handler for NMI IPIs used for stopping CPUs. */
 213         if (type == T_NMI) {
 214                  if (ipi_nmi_handler() == 0)
 215                            goto out;
 216         }
 217 #endif /* SMP */
 218
 219 #ifdef KDB
 220         if (kdb_active) {
 221                 kdb_reenter();
 222                 goto out;
 223         }
 224 #endif
 225
 226         if (type == T_RESERVED) {
 227                 trap_fatal(frame, 0);
 228                 goto out;
 229         }
 230
 231 #ifdef  HWPMC_HOOKS
 232         /*
 233          * CPU PMCs interrupt using an NMI so we check for that first.
 234          * If the HWPMC module is active, 'pmc_hook' will point to
 235          * the function to be called.  A return value of '1' from the
 236          * hook means that the NMI was handled by it and that we can
 237          * return immediately.
 238          */
 239         if (type == T_NMI && pmc_intr &&
 240             (*pmc_intr)(PCPU_GET(cpuid), frame))
 241             goto out;
 242 #endif
 243
 244         if (type == T_MCHK) {
 245                 if (!mca_intr())
 246                         trap_fatal(frame, 0);
 247                 goto out;
 248         }
 249
 250 #ifdef KDTRACE_HOOKS
 251         /*
 252          * A trap can occur while DTrace executes a probe. Before
 253          * executing the probe, DTrace blocks re-scheduling and sets
 254          * a flag in it's per-cpu flags to indicate that it doesn't
 255          * want to fault. On returning from the the probe, the no-fault
 256          * flag is cleared and finally re-scheduling is enabled.
 257          *
 258          * If the DTrace kernel module has registered a trap handler,
 259          * call it and if it returns non-zero, assume that it has
 260          * handled the trap and modified the trap frame so that this
 261          * function can return normally.
 262          */
 263         if ((type == T_PROTFLT || type == T_PAGEFLT) &&
 264             dtrace_trap_func != NULL)
 265                 if ((*dtrace_trap_func)(frame, type))
 266                         goto out;
 267 #endif
 268
 269         if ((frame->tf_eflags & PSL_I) == 0) {
 270                 /*
 271                  * Buggy application or kernel code has disabled
 272                  * interrupts and then trapped.  Enabling interrupts
 273                  * now is wrong, but it is better than running with
 274                  * interrupts disabled until they are accidentally
 275                  * enabled later.
 276                  */
 277                 if (ISPL(frame->tf_cs) == SEL_UPL || (frame->tf_eflags & PSL_VM))
 278                         uprintf(
 279                             "pid %ld (%s): trap %d with interrupts disabled\n",
 280                             (long)curproc->p_pid, curthread->td_name, type);
 281                 else if (type != T_BPTFLT && type != T_TRCTRAP &&
 282                          frame->tf_eip != (int)cpu_switch_load_gs) {
 283                         /*
 284                          * XXX not quite right, since this may be for a
 285                          * multiple fault in user mode.
 286                          */
 287                         printf("kernel trap %d with interrupts disabled\n",
 288                             type);
 289                         /*
 290                          * Page faults need interrupts disabled until later,
 291                          * and we shouldn't enable interrupts while holding
 292                          * a spin lock or if servicing an NMI.
 293                          */
 294                         if (type != T_NMI && type != T_PAGEFLT &&
 295                             td->td_md.md_spinlock_count == 0)
 296                                 enable_intr();
 297                 }
 298         }
 299         eva = 0;
 300         code = frame->tf_err;
 301         if (type == T_PAGEFLT) {
 302                 /*
 303                  * For some Cyrix CPUs, %cr2 is clobbered by
 304                  * interrupts.  This problem is worked around by using
 305                  * an interrupt gate for the pagefault handler.  We
 306                  * are finally ready to read %cr2 and then must
 307                  * reenable interrupts.
 308                  *
 309                  * If we get a page fault while in a critical section, then
 310                  * it is most likely a fatal kernel page fault.  The kernel
 311                  * is already going to panic trying to get a sleep lock to
 312                  * do the VM lookup, so just consider it a fatal trap so the
 313                  * kernel can print out a useful trap message and even get
 314                  * to the debugger.
 315                  *
 316                  * If we get a page fault while holding a non-sleepable
 317                  * lock, then it is most likely a fatal kernel page fault.
 318                  * If WITNESS is enabled, then it's going to whine about
 319                  * bogus LORs with various VM locks, so just skip to the
 320                  * fatal trap handling directly.
 321                  */
 322                 eva = rcr2();
 323                 if (td->td_critnest != 0 ||
 324                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 325                     "Kernel page fault") != 0)
 326                         trap_fatal(frame, eva);
 327                 else
 328                         enable_intr();
 329         }
 330
 331         if ((ISPL(frame->tf_cs) == SEL_UPL) ||
 332             ((frame->tf_eflags & PSL_VM) &&
 333                 !(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL))) {
 334                 /* user trap */
 335
 336                 td->td_pticks = 0;
 337                 td->td_frame = frame;
 338                 addr = frame->tf_eip;
 339                 if (td->td_ucred != p->p_ucred)
 340                         cred_update_thread(td);
 341
 342                 switch (type) {
 343                 case T_PRIVINFLT:       /* privileged instruction fault */
 344                         i = SIGILL;
 345                         ucode = ILL_PRVOPC;
 346                         break;
 347
 348                 case T_BPTFLT:          /* bpt instruction fault */
 349                 case T_TRCTRAP:         /* trace trap */
 350                         enable_intr();
 351                         frame->tf_eflags &= ~PSL_T;
 352                         i = SIGTRAP;
 353                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 354                         break;
 355
 356                 case T_ARITHTRAP:       /* arithmetic trap */
 357 #ifdef DEV_NPX
 358                         ucode = npxtrap();
 359                         if (ucode == -1)
 360                                 goto userout;
 361 #else
 362                         ucode = 0;
 363 #endif
 364                         i = SIGFPE;
 365                         break;
 366
 367                         /*
 368                          * The following two traps can happen in
 369                          * vm86 mode, and, if so, we want to handle
 370                          * them specially.
 371                          */
 372                 case T_PROTFLT:         /* general protection fault */
 373                 case T_STKFLT:          /* stack fault */
 374                         if (frame->tf_eflags & PSL_VM) {
 375                                 i = vm86_emulate((struct vm86frame *)frame);
 376                                 if (i == 0)
 377                                         goto user;
 378                                 break;
 379                         }
 380                         i = SIGBUS;
 381                         ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
 382                         break;
 383                 case T_SEGNPFLT:        /* segment not present fault */
 384                         i = SIGBUS;
 385                         ucode = BUS_ADRERR;
 386                         break;
 387                 case T_TSSFLT:          /* invalid TSS fault */
 388                         i = SIGBUS;
 389                         ucode = BUS_OBJERR;
 390                         break;
 391                 case T_DOUBLEFLT:       /* double fault */
 392                 default:
 393                         i = SIGBUS;
 394                         ucode = BUS_OBJERR;
 395                         break;
 396
 397                 case T_PAGEFLT:         /* page fault */
 398
 399                         i = trap_pfault(frame, TRUE, eva);
 400 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 401                         if (i == -2) {
 402                                 /*
 403                                  * The f00f hack workaround has triggered, so
 404                                  * treat the fault as an illegal instruction
 405                                  * (T_PRIVINFLT) instead of a page fault.
 406                                  */
 407                                 type = frame->tf_trapno = T_PRIVINFLT;
 408
 409                                 /* Proceed as in that case. */
 410                                 ucode = ILL_PRVOPC;
 411                                 i = SIGILL;
 412                                 break;
 413                         }
 414 #endif
 415                         if (i == -1)
 416                                 goto userout;
 417                         if (i == 0)
 418                                 goto user;
 419
 420                         if (i == SIGSEGV)
 421                                 ucode = SEGV_MAPERR;
 422                         else {
 423                                 if (prot_fault_translation == 0) {
 424                                         /*
 425                                          * Autodetect.
 426                                          * This check also covers the images
 427                                          * without the ABI-tag ELF note.
 428                                          */
 429                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 430                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 431                                                 i = SIGSEGV;
 432                                                 ucode = SEGV_ACCERR;
 433                                         } else {
 434                                                 i = SIGBUS;
 435                                                 ucode = BUS_PAGE_FAULT;
 436                                         }
 437                                 } else if (prot_fault_translation == 1) {
 438                                         /*
 439                                          * Always compat mode.
 440                                          */
 441                                         i = SIGBUS;
 442                                         ucode = BUS_PAGE_FAULT;
 443                                 } else {
 444                                         /*
 445                                          * Always SIGSEGV mode.
 446                                          */
 447                                         i = SIGSEGV;
 448                                         ucode = SEGV_ACCERR;
 449                                 }
 450                         }
 451                         addr = eva;
 452                         break;
 453
 454                 case T_DIVIDE:          /* integer divide fault */
 455                         ucode = FPE_INTDIV;
 456                         i = SIGFPE;
 457                         break;
 458
 459 #ifdef DEV_ISA
 460                 case T_NMI:
 461 #ifdef POWERFAIL_NMI
 462 #ifndef TIMER_FREQ
 463 #  define TIMER_FREQ 1193182
 464 #endif
 465                         if (time_second - lastalert > 10) {
 466                                 log(LOG_WARNING, "NMI: power fail\n");
 467                                 sysbeep(880, hz);
 468                                 lastalert = time_second;
 469                         }
 470                         goto userout;
 471 #else /* !POWERFAIL_NMI */
 472                         /* machine/parity/power fail/"kitchen sink" faults */
 473                         if (isa_nmi(code) == 0) {
 474 #ifdef KDB
 475                                 /*
 476                                  * NMI can be hooked up to a pushbutton
 477                                  * for debugging.
 478                                  */
 479                                 if (kdb_on_nmi) {
 480                                         printf ("NMI ... going to debugger\n");
 481                                         kdb_trap(type, 0, frame);
 482                                 }
 483 #endif /* KDB */
 484                                 goto userout;
 485                         } else if (panic_on_nmi)
 486                                 panic("NMI indicates hardware failure");
 487                         break;
 488 #endif /* POWERFAIL_NMI */
 489 #endif /* DEV_ISA */
 490
 491                 case T_OFLOW:           /* integer overflow fault */
 492                         ucode = FPE_INTOVF;
 493                         i = SIGFPE;
 494                         break;
 495
 496                 case T_BOUND:           /* bounds check fault */
 497                         ucode = FPE_FLTSUB;
 498                         i = SIGFPE;
 499                         break;
 500
 501                 case T_DNA:
 502 #ifdef DEV_NPX
 503                         KASSERT(PCB_USER_FPU(td->td_pcb),
 504                             ("kernel FPU ctx has leaked"));
 505                         /* transparent fault (due to context switch "late") */
 506                         if (npxdna())
 507                                 goto userout;
 508 #endif
 509                         uprintf("pid %d killed due to lack of floating point\n",
 510                                 p->p_pid);
 511                         i = SIGKILL;
 512                         ucode = 0;
 513                         break;
 514
 515                 case T_FPOPFLT:         /* FPU operand fetch fault */
 516                         ucode = ILL_COPROC;
 517                         i = SIGILL;
 518                         break;
 519
 520                 case T_XMMFLT:          /* SIMD floating-point exception */
 521                         ucode = 0; /* XXX */
 522                         i = SIGFPE;
 523                         break;
 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 (PCPU_GET(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                                 PCPU_GET(curpcb)->pcb_gs = 0;
 590 #if 0
 591                                 PROC_LOCK(p);
 592                                 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 (PCPU_GET(curpcb)->pcb_onfault != NULL) {
 628                                 frame->tf_eip =
 629                                     (int)PCPU_GET(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                            !(PCPU_GET(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         trapsignal(td, &ksi);
 745
 746 #ifdef DEBUG
 747         if (type <= MAX_TRAP_MSG) {
 748                 uprintf("fatal process exception: %s",
 749                         trap_msg[type]);
 750                 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
 751                         uprintf(", fault VA = 0x%lx", (u_long)eva);
 752                 uprintf("\n");
 753         }
 754 #endif
 755
 756 user:
 757         userret(td, frame);
 758         mtx_assert(&Giant, MA_NOTOWNED);
 759         KASSERT(PCB_USER_FPU(td->td_pcb),
 760             ("Return from trap with kernel FPU ctx leaked"));
 761 userout:
 762 out:
 763         return;
 764 }
 765
 766 static int
 767 trap_pfault(frame, usermode, eva)
 768         struct trapframe *frame;
 769         int usermode;
 770         vm_offset_t eva;
 771 {
 772         vm_offset_t va;
 773         struct vmspace *vm = NULL;
 774         vm_map_t map;
 775         int rv = 0;
 776         vm_prot_t ftype;
 777         struct thread *td = curthread;
 778         struct proc *p = td->td_proc;
 779
 780         va = trunc_page(eva);
 781         if (va >= KERNBASE) {
 782                 /*
 783                  * Don't allow user-mode faults in kernel address space.
 784                  * An exception:  if the faulting address is the invalid
 785                  * instruction entry in the IDT, then the Intel Pentium
 786                  * F00F bug workaround was triggered, and we need to
 787                  * treat it is as an illegal instruction, and not a page
 788                  * fault.
 789                  */
 790 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 791                 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
 792                         return -2;
 793 #endif
 794                 if (usermode)
 795                         goto nogo;
 796
 797                 map = kernel_map;
 798         } else {
 799                 /*
 800                  * This is a fault on non-kernel virtual memory.
 801                  * vm is initialized above to NULL. If curproc is NULL
 802                  * or curproc->p_vmspace is NULL the fault is fatal.
 803                  */
 804                 if (p != NULL)
 805                         vm = p->p_vmspace;
 806
 807                 if (vm == NULL)
 808                         goto nogo;
 809
 810                 map = &vm->vm_map;
 811         }
 812
 813         /*
 814          * PGEX_I is defined only if the execute disable bit capability is
 815          * supported and enabled.
 816          */
 817         if (frame->tf_err & PGEX_W)
 818                 ftype = VM_PROT_WRITE;
 819 #ifdef PAE
 820         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 821                 ftype = VM_PROT_EXECUTE;
 822 #endif
 823         else
 824                 ftype = VM_PROT_READ;
 825
 826         if (map != kernel_map) {
 827                 /*
 828                  * Keep swapout from messing with us during this
 829                  *      critical time.
 830                  */
 831                 PROC_LOCK(p);
 832                 ++p->p_lock;
 833                 PROC_UNLOCK(p);
 834
 835                 /* Fault in the user page: */
 836                 rv = vm_fault(map, va, ftype,
 837                               (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
 838                                                       : VM_FAULT_NORMAL);
 839
 840                 PROC_LOCK(p);
 841                 --p->p_lock;
 842                 PROC_UNLOCK(p);
 843         } else {
 844                 /*
 845                  * Don't have to worry about process locking or stacks in the
 846                  * kernel.
 847                  */
 848                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 849         }
 850         if (rv == KERN_SUCCESS)
 851                 return (0);
 852 nogo:
 853         if (!usermode) {
 854                 if (td->td_intr_nesting_level == 0 &&
 855                     PCPU_GET(curpcb)->pcb_onfault != NULL) {
 856                         frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault;
 857                         return (0);
 858                 }
 859                 trap_fatal(frame, eva);
 860                 return (-1);
 861         }
 862
 863         return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 864 }
 865
 866 static void
 867 trap_fatal(frame, eva)
 868         struct trapframe *frame;
 869         vm_offset_t eva;
 870 {
 871         int code, ss, esp;
 872         u_int type;
 873         struct soft_segment_descriptor softseg;
 874         char *msg;
 875
 876         code = frame->tf_err;
 877         type = frame->tf_trapno;
 878         sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
 879
 880         if (type <= MAX_TRAP_MSG)
 881                 msg = trap_msg[type];
 882         else
 883                 msg = "UNKNOWN";
 884         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 885             frame->tf_eflags & PSL_VM ? "vm86" :
 886             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
 887 #ifdef SMP
 888         /* two separate prints in case of a trap on an unmapped page */
 889         printf("cpuid = %d; ", PCPU_GET(cpuid));
 890         printf("apic id = %02x\n", PCPU_GET(apic_id));
 891 #endif
 892         if (type == T_PAGEFLT) {
 893                 printf("fault virtual address   = 0x%x\n", eva);
 894                 printf("fault code              = %s %s, %s\n",
 895                         code & PGEX_U ? "user" : "supervisor",
 896                         code & PGEX_W ? "write" : "read",
 897                         code & PGEX_P ? "protection violation" : "page not present");
 898         }
 899         printf("instruction pointer     = 0x%x:0x%x\n",
 900                frame->tf_cs & 0xffff, frame->tf_eip);
 901         if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
 902                 ss = frame->tf_ss & 0xffff;
 903                 esp = frame->tf_esp;
 904         } else {
 905                 ss = GSEL(GDATA_SEL, SEL_KPL);
 906                 esp = (int)&frame->tf_esp;
 907         }
 908         printf("stack pointer           = 0x%x:0x%x\n", ss, esp);
 909         printf("frame pointer           = 0x%x:0x%x\n", ss, frame->tf_ebp);
 910         printf("code segment            = base 0x%x, limit 0x%x, type 0x%x\n",
 911                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 912         printf("                        = DPL %d, pres %d, def32 %d, gran %d\n",
 913                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
 914                softseg.ssd_gran);
 915         printf("processor eflags        = ");
 916         if (frame->tf_eflags & PSL_T)
 917                 printf("trace trap, ");
 918         if (frame->tf_eflags & PSL_I)
 919                 printf("interrupt enabled, ");
 920         if (frame->tf_eflags & PSL_NT)
 921                 printf("nested task, ");
 922         if (frame->tf_eflags & PSL_RF)
 923                 printf("resume, ");
 924         if (frame->tf_eflags & PSL_VM)
 925                 printf("vm86, ");
 926         printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
 927         printf("current process         = ");
 928         if (curproc) {
 929                 printf("%lu (%s)\n", (u_long)curproc->p_pid, curthread->td_name);
 930         } else {
 931                 printf("Idle\n");
 932         }
 933
 934 #ifdef KDB
 935         if (debugger_on_panic || kdb_active) {
 936                 frame->tf_err = eva;    /* smuggle fault address to ddb */
 937                 if (kdb_trap(type, 0, frame)) {
 938                         frame->tf_err = code;   /* restore error code */
 939                         return;
 940                 }
 941                 frame->tf_err = code;           /* restore error code */
 942         }
 943 #endif
 944         printf("trap number             = %d\n", type);
 945         if (type <= MAX_TRAP_MSG)
 946                 panic("%s", trap_msg[type]);
 947         else
 948                 panic("unknown/reserved trap");
 949 }
 950
 951 /*
 952  * Double fault handler. Called when a fault occurs while writing
 953  * a frame for a trap/exception onto the stack. This usually occurs
 954  * when the stack overflows (such is the case with infinite recursion,
 955  * for example).
 956  *
 957  * XXX Note that the current PTD gets replaced by IdlePTD when the
 958  * task switch occurs. This means that the stack that was active at
 959  * the time of the double fault is not available at <kstack> unless
 960  * the machine was idle when the double fault occurred. The downside
 961  * of this is that "trace <ebp>" in ddb won't work.
 962  */
 963 void
 964 dblfault_handler()
 965 {
 966 #ifdef KDTRACE_HOOKS
 967         if (dtrace_doubletrap_func != NULL)
 968                 (*dtrace_doubletrap_func)();
 969 #endif
 970         printf("\nFatal double fault:\n");
 971         printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip));
 972         printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp));
 973         printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp));
 974 #ifdef SMP
 975         /* two separate prints in case of a trap on an unmapped page */
 976         printf("cpuid = %d; ", PCPU_GET(cpuid));
 977         printf("apic id = %02x\n", PCPU_GET(apic_id));
 978 #endif
 979         panic("double fault");
 980 }
 981
 982 int
 983 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
 984 {
 985         struct proc *p;
 986         struct trapframe *frame;
 987         caddr_t params;
 988         int error;
 989
 990         p = td->td_proc;
 991         frame = td->td_frame;
 992
 993         params = (caddr_t)frame->tf_esp + sizeof(int);
 994         sa->code = frame->tf_eax;
 995
 996         /*
 997          * Need to check if this is a 32 bit or 64 bit syscall.
 998          */
 999         if (sa->code == SYS_syscall) {
1000                 /*
1001                  * Code is first argument, followed by actual args.
1002                  */
1003                 sa->code = fuword(params);
1004                 params += sizeof(int);
1005         } else if (sa->code == SYS___syscall) {
1006                 /*
1007                  * Like syscall, but code is a quad, so as to maintain
1008                  * quad alignment for the rest of the arguments.
1009                  */
1010                 sa->code = fuword(params);
1011                 params += sizeof(quad_t);
1012         }
1013
1014         if (p->p_sysent->sv_mask)
1015                 sa->code &= p->p_sysent->sv_mask;
1016         if (sa->code >= p->p_sysent->sv_size)
1017                 sa->callp = &p->p_sysent->sv_table[0];
1018         else
1019                 sa->callp = &p->p_sysent->sv_table[sa->code];
1020         sa->narg = sa->callp->sy_narg;
1021
1022         if (params != NULL && sa->narg != 0)
1023                 error = copyin(params, (caddr_t)sa->args,
1024                     (u_int)(sa->narg * sizeof(int)));
1025         else
1026                 error = 0;
1027
1028         if (error == 0) {
1029                 td->td_retval[0] = 0;
1030                 td->td_retval[1] = frame->tf_edx;
1031         }
1032
1033         return (error);
1034 }
1035
1036 /*
1037  *      syscall -       system call request C handler
1038  *
1039  *      A system call is essentially treated as a trap.
1040  */
1041 void
1042 syscall(struct trapframe *frame)
1043 {
1044         struct thread *td;
1045         struct syscall_args sa;
1046         register_t orig_tf_eflags;
1047         int error;
1048         ksiginfo_t ksi;
1049
1050 #ifdef DIAGNOSTIC
1051         if (ISPL(frame->tf_cs) != SEL_UPL) {
1052                 panic("syscall");
1053                 /* NOT REACHED */
1054         }
1055 #endif
1056         orig_tf_eflags = frame->tf_eflags;
1057
1058         td = curthread;
1059         td->td_frame = frame;
1060
1061         error = syscallenter(td, &sa);
1062
1063         /*
1064          * Traced syscall.
1065          */
1066         if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1067                 frame->tf_eflags &= ~PSL_T;
1068                 ksiginfo_init_trap(&ksi);
1069                 ksi.ksi_signo = SIGTRAP;
1070                 ksi.ksi_code = TRAP_TRACE;
1071                 ksi.ksi_addr = (void *)frame->tf_eip;
1072                 trapsignal(td, &ksi);
1073         }
1074
1075         KASSERT(PCB_USER_FPU(td->td_pcb),
1076             ("System call %s returning with kernel FPU ctx leaked",
1077              syscallname(td->td_proc, sa.code)));
1078         KASSERT(td->td_pcb->pcb_save == &td->td_pcb->pcb_user_save,
1079             ("System call %s returning with mangled pcb_save",
1080              syscallname(td->td_proc, sa.code)));
1081
1082         syscallret(td, error, &sa);
1083 }