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