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            33
 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         "",                                     /* 31 unused (reserved) */
 173         "DTrace pid return trap",               /* 32 T_DTRACE_RET */
 174         "DTrace fasttrap probe trap",           /* 33 T_DTRACE_PROBE */
 175
 176 };
 177
 178 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 179 extern int has_f00f_bug;
 180 #endif
 181
 182 #ifdef KDB
 183 static int kdb_on_nmi = 1;
 184 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
 185         &kdb_on_nmi, 0, "Go to KDB on NMI");
 186 TUNABLE_INT("machdep.kdb_on_nmi", &kdb_on_nmi);
 187 #endif
 188 static int panic_on_nmi = 1;
 189 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
 190         &panic_on_nmi, 0, "Panic on NMI");
 191 TUNABLE_INT("machdep.panic_on_nmi", &panic_on_nmi);
 192 static int prot_fault_translation = 0;
 193 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
 194         &prot_fault_translation, 0, "Select signal to deliver on protection fault");
 195
 196 /*
 197  * Exception, fault, and trap interface to the FreeBSD kernel.
 198  * This common code is called from assembly language IDT gate entry
 199  * routines that prepare a suitable stack frame, and restore this
 200  * frame after the exception has been processed.
 201  */
 202
 203 void
 204 trap(struct trapframe *frame)
 205 {
 206         struct thread *td = curthread;
 207         struct proc *p = td->td_proc;
 208         int i = 0, ucode = 0, code;
 209         u_int type;
 210         register_t addr = 0;
 211         vm_offset_t eva;
 212         ksiginfo_t ksi;
 213 #ifdef POWERFAIL_NMI
 214         static int lastalert = 0;
 215 #endif
 216
 217         PCPU_INC(cnt.v_trap);
 218         type = frame->tf_trapno;
 219
 220 #ifdef SMP
 221         /* Handler for NMI IPIs used for stopping CPUs. */
 222         if (type == T_NMI) {
 223                  if (ipi_nmi_handler() == 0)
 224                            goto out;
 225         }
 226 #endif /* SMP */
 227
 228 #ifdef KDB
 229         if (kdb_active) {
 230                 kdb_reenter();
 231                 goto out;
 232         }
 233 #endif
 234
 235         if (type == T_RESERVED) {
 236                 trap_fatal(frame, 0);
 237                 goto out;
 238         }
 239
 240 #ifdef  HWPMC_HOOKS
 241         /*
 242          * CPU PMCs interrupt using an NMI so we check for that first.
 243          * If the HWPMC module is active, 'pmc_hook' will point to
 244          * the function to be called.  A return value of '1' from the
 245          * hook means that the NMI was handled by it and that we can
 246          * return immediately.
 247          */
 248         if (type == T_NMI && pmc_intr &&
 249             (*pmc_intr)(PCPU_GET(cpuid), frame))
 250             goto out;
 251 #endif
 252
 253         if (type == T_MCHK) {
 254                 if (!mca_intr())
 255                         trap_fatal(frame, 0);
 256                 goto out;
 257         }
 258
 259 #ifdef KDTRACE_HOOKS
 260         /*
 261          * A trap can occur while DTrace executes a probe. Before
 262          * executing the probe, DTrace blocks re-scheduling and sets
 263          * a flag in it's per-cpu flags to indicate that it doesn't
 264          * want to fault. On returning from the probe, the no-fault
 265          * flag is cleared and finally re-scheduling is enabled.
 266          *
 267          * If the DTrace kernel module has registered a trap handler,
 268          * call it and if it returns non-zero, assume that it has
 269          * handled the trap and modified the trap frame so that this
 270          * function can return normally.
 271          */
 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         if ((type == T_PROTFLT || type == T_PAGEFLT) &&
 291             dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
 292                 goto out;
 293 #endif
 294
 295         if ((frame->tf_eflags & PSL_I) == 0) {
 296                 /*
 297                  * Buggy application or kernel code has disabled
 298                  * interrupts and then trapped.  Enabling interrupts
 299                  * now is wrong, but it is better than running with
 300                  * interrupts disabled until they are accidentally
 301                  * enabled later.
 302                  */
 303                 if (ISPL(frame->tf_cs) == SEL_UPL || (frame->tf_eflags & PSL_VM))
 304                         uprintf(
 305                             "pid %ld (%s): trap %d with interrupts disabled\n",
 306                             (long)curproc->p_pid, curthread->td_name, type);
 307                 else if (type != T_BPTFLT && type != T_TRCTRAP &&
 308                          frame->tf_eip != (int)cpu_switch_load_gs) {
 309                         /*
 310                          * XXX not quite right, since this may be for a
 311                          * multiple fault in user mode.
 312                          */
 313                         printf("kernel trap %d with interrupts disabled\n",
 314                             type);
 315                         /*
 316                          * Page faults need interrupts disabled until later,
 317                          * and we shouldn't enable interrupts while holding
 318                          * a spin lock or if servicing an NMI.
 319                          */
 320                         if (type != T_NMI && type != T_PAGEFLT &&
 321                             td->td_md.md_spinlock_count == 0)
 322                                 enable_intr();
 323                 }
 324         }
 325         eva = 0;
 326         code = frame->tf_err;
 327         if (type == T_PAGEFLT) {
 328                 /*
 329                  * For some Cyrix CPUs, %cr2 is clobbered by
 330                  * interrupts.  This problem is worked around by using
 331                  * an interrupt gate for the pagefault handler.  We
 332                  * are finally ready to read %cr2 and then must
 333                  * reenable interrupts.
 334                  *
 335                  * If we get a page fault while in a critical section, then
 336                  * it is most likely a fatal kernel page fault.  The kernel
 337                  * is already going to panic trying to get a sleep lock to
 338                  * do the VM lookup, so just consider it a fatal trap so the
 339                  * kernel can print out a useful trap message and even get
 340                  * to the debugger.
 341                  *
 342                  * If we get a page fault while holding a non-sleepable
 343                  * lock, then it is most likely a fatal kernel page fault.
 344                  * If WITNESS is enabled, then it's going to whine about
 345                  * bogus LORs with various VM locks, so just skip to the
 346                  * fatal trap handling directly.
 347                  */
 348                 eva = rcr2();
 349                 if (td->td_critnest != 0 ||
 350                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 351                     "Kernel page fault") != 0)
 352                         trap_fatal(frame, eva);
 353                 else
 354                         enable_intr();
 355         }
 356
 357         if ((ISPL(frame->tf_cs) == SEL_UPL) ||
 358             ((frame->tf_eflags & PSL_VM) &&
 359                 !(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL))) {
 360                 /* user trap */
 361
 362                 td->td_pticks = 0;
 363                 td->td_frame = frame;
 364                 addr = frame->tf_eip;
 365                 if (td->td_ucred != p->p_ucred)
 366                         cred_update_thread(td);
 367
 368                 switch (type) {
 369                 case T_PRIVINFLT:       /* privileged instruction fault */
 370                         i = SIGILL;
 371                         ucode = ILL_PRVOPC;
 372                         break;
 373
 374                 case T_BPTFLT:          /* bpt instruction fault */
 375                 case T_TRCTRAP:         /* trace trap */
 376                         enable_intr();
 377                         frame->tf_eflags &= ~PSL_T;
 378                         i = SIGTRAP;
 379                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 380                         break;
 381
 382                 case T_ARITHTRAP:       /* arithmetic trap */
 383 #ifdef DEV_NPX
 384                         ucode = npxtrap();
 385                         if (ucode == -1)
 386                                 goto userout;
 387 #else
 388                         ucode = 0;
 389 #endif
 390                         i = SIGFPE;
 391                         break;
 392
 393                         /*
 394                          * The following two traps can happen in
 395                          * vm86 mode, and, if so, we want to handle
 396                          * them specially.
 397                          */
 398                 case T_PROTFLT:         /* general protection fault */
 399                 case T_STKFLT:          /* stack fault */
 400                         if (frame->tf_eflags & PSL_VM) {
 401                                 i = vm86_emulate((struct vm86frame *)frame);
 402                                 if (i == 0)
 403                                         goto user;
 404                                 break;
 405                         }
 406                         i = SIGBUS;
 407                         ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
 408                         break;
 409                 case T_SEGNPFLT:        /* segment not present fault */
 410                         i = SIGBUS;
 411                         ucode = BUS_ADRERR;
 412                         break;
 413                 case T_TSSFLT:          /* invalid TSS fault */
 414                         i = SIGBUS;
 415                         ucode = BUS_OBJERR;
 416                         break;
 417                 case T_DOUBLEFLT:       /* double fault */
 418                 default:
 419                         i = SIGBUS;
 420                         ucode = BUS_OBJERR;
 421                         break;
 422
 423                 case T_PAGEFLT:         /* page fault */
 424
 425                         i = trap_pfault(frame, TRUE, eva);
 426 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 427                         if (i == -2) {
 428                                 /*
 429                                  * The f00f hack workaround has triggered, so
 430                                  * treat the fault as an illegal instruction
 431                                  * (T_PRIVINFLT) instead of a page fault.
 432                                  */
 433                                 type = frame->tf_trapno = T_PRIVINFLT;
 434
 435                                 /* Proceed as in that case. */
 436                                 ucode = ILL_PRVOPC;
 437                                 i = SIGILL;
 438                                 break;
 439                         }
 440 #endif
 441                         if (i == -1)
 442                                 goto userout;
 443                         if (i == 0)
 444                                 goto user;
 445
 446                         if (i == SIGSEGV)
 447                                 ucode = SEGV_MAPERR;
 448                         else {
 449                                 if (prot_fault_translation == 0) {
 450                                         /*
 451                                          * Autodetect.
 452                                          * This check also covers the images
 453                                          * without the ABI-tag ELF note.
 454                                          */
 455                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 456                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 457                                                 i = SIGSEGV;
 458                                                 ucode = SEGV_ACCERR;
 459                                         } else {
 460                                                 i = SIGBUS;
 461                                                 ucode = BUS_PAGE_FAULT;
 462                                         }
 463                                 } else if (prot_fault_translation == 1) {
 464                                         /*
 465                                          * Always compat mode.
 466                                          */
 467                                         i = SIGBUS;
 468                                         ucode = BUS_PAGE_FAULT;
 469                                 } else {
 470                                         /*
 471                                          * Always SIGSEGV mode.
 472                                          */
 473                                         i = SIGSEGV;
 474                                         ucode = SEGV_ACCERR;
 475                                 }
 476                         }
 477                         addr = eva;
 478                         break;
 479
 480                 case T_DIVIDE:          /* integer divide fault */
 481                         ucode = FPE_INTDIV;
 482                         i = SIGFPE;
 483                         break;
 484
 485 #ifdef DEV_ISA
 486                 case T_NMI:
 487 #ifdef POWERFAIL_NMI
 488 #ifndef TIMER_FREQ
 489 #  define TIMER_FREQ 1193182
 490 #endif
 491                         if (time_second - lastalert > 10) {
 492                                 log(LOG_WARNING, "NMI: power fail\n");
 493                                 sysbeep(880, hz);
 494                                 lastalert = time_second;
 495                         }
 496                         goto userout;
 497 #else /* !POWERFAIL_NMI */
 498                         /* machine/parity/power fail/"kitchen sink" faults */
 499                         if (isa_nmi(code) == 0) {
 500 #ifdef KDB
 501                                 /*
 502                                  * NMI can be hooked up to a pushbutton
 503                                  * for debugging.
 504                                  */
 505                                 if (kdb_on_nmi) {
 506                                         printf ("NMI ... going to debugger\n");
 507                                         kdb_trap(type, 0, frame);
 508                                 }
 509 #endif /* KDB */
 510                                 goto userout;
 511                         } else if (panic_on_nmi)
 512                                 panic("NMI indicates hardware failure");
 513                         break;
 514 #endif /* POWERFAIL_NMI */
 515 #endif /* DEV_ISA */
 516
 517                 case T_OFLOW:           /* integer overflow fault */
 518                         ucode = FPE_INTOVF;
 519                         i = SIGFPE;
 520                         break;
 521
 522                 case T_BOUND:           /* bounds check fault */
 523                         ucode = FPE_FLTSUB;
 524                         i = SIGFPE;
 525                         break;
 526
 527                 case T_DNA:
 528 #ifdef DEV_NPX
 529                         KASSERT(PCB_USER_FPU(td->td_pcb),
 530                             ("kernel FPU ctx has leaked"));
 531                         /* transparent fault (due to context switch "late") */
 532                         if (npxdna())
 533                                 goto userout;
 534 #endif
 535                         uprintf("pid %d killed due to lack of floating point\n",
 536                                 p->p_pid);
 537                         i = SIGKILL;
 538                         ucode = 0;
 539                         break;
 540
 541                 case T_FPOPFLT:         /* FPU operand fetch fault */
 542                         ucode = ILL_COPROC;
 543                         i = SIGILL;
 544                         break;
 545
 546                 case T_XMMFLT:          /* SIMD floating-point exception */
 547                         ucode = 0; /* XXX */
 548                         i = SIGFPE;
 549                         break;
 550                 }
 551         } else {
 552                 /* kernel trap */
 553
 554                 KASSERT(cold || td->td_ucred != NULL,
 555                     ("kernel trap doesn't have ucred"));
 556                 switch (type) {
 557                 case T_PAGEFLT:                 /* page fault */
 558                         (void) trap_pfault(frame, FALSE, eva);
 559                         goto out;
 560
 561                 case T_DNA:
 562 #ifdef DEV_NPX
 563                         KASSERT(!PCB_USER_FPU(td->td_pcb),
 564                             ("Unregistered use of FPU in kernel"));
 565                         if (npxdna())
 566                                 goto out;
 567 #endif
 568                         break;
 569
 570                 case T_ARITHTRAP:       /* arithmetic trap */
 571                 case T_XMMFLT:          /* SIMD floating-point exception */
 572                 case T_FPOPFLT:         /* FPU operand fetch fault */
 573                         /*
 574                          * XXXKIB for now disable any FPU traps in kernel
 575                          * handler registration seems to be overkill
 576                          */
 577                         trap_fatal(frame, 0);
 578                         goto out;
 579
 580                         /*
 581                          * The following two traps can happen in
 582                          * vm86 mode, and, if so, we want to handle
 583                          * them specially.
 584                          */
 585                 case T_PROTFLT:         /* general protection fault */
 586                 case T_STKFLT:          /* stack fault */
 587                         if (frame->tf_eflags & PSL_VM) {
 588                                 i = vm86_emulate((struct vm86frame *)frame);
 589                                 if (i != 0)
 590                                         /*
 591                                          * returns to original process
 592                                          */
 593                                         vm86_trap((struct vm86frame *)frame);
 594                                 goto out;
 595                         }
 596                         if (type == T_STKFLT)
 597                                 break;
 598
 599                         /* FALL THROUGH */
 600
 601                 case T_SEGNPFLT:        /* segment not present fault */
 602                         if (PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL)
 603                                 break;
 604
 605                         /*
 606                          * Invalid %fs's and %gs's can be created using
 607                          * procfs or PT_SETREGS or by invalidating the
 608                          * underlying LDT entry.  This causes a fault
 609                          * in kernel mode when the kernel attempts to
 610                          * switch contexts.  Lose the bad context
 611                          * (XXX) so that we can continue, and generate
 612                          * a signal.
 613                          */
 614                         if (frame->tf_eip == (int)cpu_switch_load_gs) {
 615                                 PCPU_GET(curpcb)->pcb_gs = 0;
 616 #if 0
 617                                 PROC_LOCK(p);
 618                                 kern_psignal(p, SIGBUS);
 619                                 PROC_UNLOCK(p);
 620 #endif
 621                                 goto out;
 622                         }
 623
 624                         if (td->td_intr_nesting_level != 0)
 625                                 break;
 626
 627                         /*
 628                          * Invalid segment selectors and out of bounds
 629                          * %eip's and %esp's can be set up in user mode.
 630                          * This causes a fault in kernel mode when the
 631                          * kernel tries to return to user mode.  We want
 632                          * to get this fault so that we can fix the
 633                          * problem here and not have to check all the
 634                          * selectors and pointers when the user changes
 635                          * them.
 636                          */
 637                         if (frame->tf_eip == (int)doreti_iret) {
 638                                 frame->tf_eip = (int)doreti_iret_fault;
 639                                 goto out;
 640                         }
 641                         if (frame->tf_eip == (int)doreti_popl_ds) {
 642                                 frame->tf_eip = (int)doreti_popl_ds_fault;
 643                                 goto out;
 644                         }
 645                         if (frame->tf_eip == (int)doreti_popl_es) {
 646                                 frame->tf_eip = (int)doreti_popl_es_fault;
 647                                 goto out;
 648                         }
 649                         if (frame->tf_eip == (int)doreti_popl_fs) {
 650                                 frame->tf_eip = (int)doreti_popl_fs_fault;
 651                                 goto out;
 652                         }
 653                         if (PCPU_GET(curpcb)->pcb_onfault != NULL) {
 654                                 frame->tf_eip =
 655                                     (int)PCPU_GET(curpcb)->pcb_onfault;
 656                                 goto out;
 657                         }
 658                         break;
 659
 660                 case T_TSSFLT:
 661                         /*
 662                          * PSL_NT can be set in user mode and isn't cleared
 663                          * automatically when the kernel is entered.  This
 664                          * causes a TSS fault when the kernel attempts to
 665                          * `iret' because the TSS link is uninitialized.  We
 666                          * want to get this fault so that we can fix the
 667                          * problem here and not every time the kernel is
 668                          * entered.
 669                          */
 670                         if (frame->tf_eflags & PSL_NT) {
 671                                 frame->tf_eflags &= ~PSL_NT;
 672                                 goto out;
 673                         }
 674                         break;
 675
 676                 case T_TRCTRAP:  /* trace trap */
 677                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall)) {
 678                                 /*
 679                                  * We've just entered system mode via the
 680                                  * syscall lcall.  Continue single stepping
 681                                  * silently until the syscall handler has
 682                                  * saved the flags.
 683                                  */
 684                                 goto out;
 685                         }
 686                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall) + 1) {
 687                                 /*
 688                                  * The syscall handler has now saved the
 689                                  * flags.  Stop single stepping it.
 690                                  */
 691                                 frame->tf_eflags &= ~PSL_T;
 692                                 goto out;
 693                         }
 694                         /*
 695                          * Ignore debug register trace traps due to
 696                          * accesses in the user's address space, which
 697                          * can happen under several conditions such as
 698                          * if a user sets a watchpoint on a buffer and
 699                          * then passes that buffer to a system call.
 700                          * We still want to get TRCTRAPS for addresses
 701                          * in kernel space because that is useful when
 702                          * debugging the kernel.
 703                          */
 704                         if (user_dbreg_trap() &&
 705                            !(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL)) {
 706                                 /*
 707                                  * Reset breakpoint bits because the
 708                                  * processor doesn't
 709                                  */
 710                                 load_dr6(rdr6() & 0xfffffff0);
 711                                 goto out;
 712                         }
 713                         /*
 714                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 715                          */
 716                 case T_BPTFLT:
 717                         /*
 718                          * If KDB is enabled, let it handle the debugger trap.
 719                          * Otherwise, debugger traps "can't happen".
 720                          */
 721 #ifdef KDB
 722                         if (kdb_trap(type, 0, frame))
 723                                 goto out;
 724 #endif
 725                         break;
 726
 727 #ifdef DEV_ISA
 728                 case T_NMI:
 729 #ifdef POWERFAIL_NMI
 730                         if (time_second - lastalert > 10) {
 731                                 log(LOG_WARNING, "NMI: power fail\n");
 732                                 sysbeep(880, hz);
 733                                 lastalert = time_second;
 734                         }
 735                         goto out;
 736 #else /* !POWERFAIL_NMI */
 737                         /* machine/parity/power fail/"kitchen sink" faults */
 738                         if (isa_nmi(code) == 0) {
 739 #ifdef KDB
 740                                 /*
 741                                  * NMI can be hooked up to a pushbutton
 742                                  * for debugging.
 743                                  */
 744                                 if (kdb_on_nmi) {
 745                                         printf ("NMI ... going to debugger\n");
 746                                         kdb_trap(type, 0, frame);
 747                                 }
 748 #endif /* KDB */
 749                                 goto out;
 750                         } else if (panic_on_nmi == 0)
 751                                 goto out;
 752                         /* FALLTHROUGH */
 753 #endif /* POWERFAIL_NMI */
 754 #endif /* DEV_ISA */
 755                 }
 756
 757                 trap_fatal(frame, eva);
 758                 goto out;
 759         }
 760
 761         /* Translate fault for emulators (e.g. Linux) */
 762         if (*p->p_sysent->sv_transtrap)
 763                 i = (*p->p_sysent->sv_transtrap)(i, type);
 764
 765         ksiginfo_init_trap(&ksi);
 766         ksi.ksi_signo = i;
 767         ksi.ksi_code = ucode;
 768         ksi.ksi_addr = (void *)addr;
 769         ksi.ksi_trapno = type;
 770         trapsignal(td, &ksi);
 771
 772 #ifdef DEBUG
 773         if (type <= MAX_TRAP_MSG) {
 774                 uprintf("fatal process exception: %s",
 775                         trap_msg[type]);
 776                 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
 777                         uprintf(", fault VA = 0x%lx", (u_long)eva);
 778                 uprintf("\n");
 779         }
 780 #endif
 781
 782 user:
 783         userret(td, frame);
 784         mtx_assert(&Giant, MA_NOTOWNED);
 785         KASSERT(PCB_USER_FPU(td->td_pcb),
 786             ("Return from trap with kernel FPU ctx leaked"));
 787 userout:
 788 out:
 789         return;
 790 }
 791
 792 static int
 793 trap_pfault(frame, usermode, eva)
 794         struct trapframe *frame;
 795         int usermode;
 796         vm_offset_t eva;
 797 {
 798         vm_offset_t va;
 799         struct vmspace *vm = NULL;
 800         vm_map_t map;
 801         int rv = 0;
 802         vm_prot_t ftype;
 803         struct thread *td = curthread;
 804         struct proc *p = td->td_proc;
 805
 806         va = trunc_page(eva);
 807         if (va >= KERNBASE) {
 808                 /*
 809                  * Don't allow user-mode faults in kernel address space.
 810                  * An exception:  if the faulting address is the invalid
 811                  * instruction entry in the IDT, then the Intel Pentium
 812                  * F00F bug workaround was triggered, and we need to
 813                  * treat it is as an illegal instruction, and not a page
 814                  * fault.
 815                  */
 816 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 817                 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
 818                         return -2;
 819 #endif
 820                 if (usermode)
 821                         goto nogo;
 822
 823                 map = kernel_map;
 824         } else {
 825                 /*
 826                  * This is a fault on non-kernel virtual memory.
 827                  * vm is initialized above to NULL. If curproc is NULL
 828                  * or curproc->p_vmspace is NULL the fault is fatal.
 829                  */
 830                 if (p != NULL)
 831                         vm = p->p_vmspace;
 832
 833                 if (vm == NULL)
 834                         goto nogo;
 835
 836                 map = &vm->vm_map;
 837                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 838                     PCPU_GET(curpcb)->pcb_onfault == NULL)) {
 839                         trap_fatal(frame, eva);
 840                         return (-1);
 841                 }
 842         }
 843
 844         /*
 845          * PGEX_I is defined only if the execute disable bit capability is
 846          * supported and enabled.
 847          */
 848         if (frame->tf_err & PGEX_W)
 849                 ftype = VM_PROT_WRITE;
 850 #ifdef PAE
 851         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 852                 ftype = VM_PROT_EXECUTE;
 853 #endif
 854         else
 855                 ftype = VM_PROT_READ;
 856
 857         if (map != kernel_map) {
 858                 /*
 859                  * Keep swapout from messing with us during this
 860                  *      critical time.
 861                  */
 862                 PROC_LOCK(p);
 863                 ++p->p_lock;
 864                 PROC_UNLOCK(p);
 865
 866                 /* Fault in the user page: */
 867                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 868
 869                 PROC_LOCK(p);
 870                 --p->p_lock;
 871                 PROC_UNLOCK(p);
 872         } else {
 873                 /*
 874                  * Don't have to worry about process locking or stacks in the
 875                  * kernel.
 876                  */
 877                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 878         }
 879         if (rv == KERN_SUCCESS)
 880                 return (0);
 881 nogo:
 882         if (!usermode) {
 883                 if (td->td_intr_nesting_level == 0 &&
 884                     PCPU_GET(curpcb)->pcb_onfault != NULL) {
 885                         frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault;
 886                         return (0);
 887                 }
 888                 trap_fatal(frame, eva);
 889                 return (-1);
 890         }
 891
 892         return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 893 }
 894
 895 static void
 896 trap_fatal(frame, eva)
 897         struct trapframe *frame;
 898         vm_offset_t eva;
 899 {
 900         int code, ss, esp;
 901         u_int type;
 902         struct soft_segment_descriptor softseg;
 903         char *msg;
 904
 905         code = frame->tf_err;
 906         type = frame->tf_trapno;
 907         sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
 908
 909         if (type <= MAX_TRAP_MSG)
 910                 msg = trap_msg[type];
 911         else
 912                 msg = "UNKNOWN";
 913         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 914             frame->tf_eflags & PSL_VM ? "vm86" :
 915             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
 916 #ifdef SMP
 917         /* two separate prints in case of a trap on an unmapped page */
 918         printf("cpuid = %d; ", PCPU_GET(cpuid));
 919         printf("apic id = %02x\n", PCPU_GET(apic_id));
 920 #endif
 921         if (type == T_PAGEFLT) {
 922                 printf("fault virtual address   = 0x%x\n", eva);
 923                 printf("fault code              = %s %s, %s\n",
 924                         code & PGEX_U ? "user" : "supervisor",
 925                         code & PGEX_W ? "write" : "read",
 926                         code & PGEX_P ? "protection violation" : "page not present");
 927         }
 928         printf("instruction pointer     = 0x%x:0x%x\n",
 929                frame->tf_cs & 0xffff, frame->tf_eip);
 930         if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
 931                 ss = frame->tf_ss & 0xffff;
 932                 esp = frame->tf_esp;
 933         } else {
 934                 ss = GSEL(GDATA_SEL, SEL_KPL);
 935                 esp = (int)&frame->tf_esp;
 936         }
 937         printf("stack pointer           = 0x%x:0x%x\n", ss, esp);
 938         printf("frame pointer           = 0x%x:0x%x\n", ss, frame->tf_ebp);
 939         printf("code segment            = base 0x%x, limit 0x%x, type 0x%x\n",
 940                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 941         printf("                        = DPL %d, pres %d, def32 %d, gran %d\n",
 942                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
 943                softseg.ssd_gran);
 944         printf("processor eflags        = ");
 945         if (frame->tf_eflags & PSL_T)
 946                 printf("trace trap, ");
 947         if (frame->tf_eflags & PSL_I)
 948                 printf("interrupt enabled, ");
 949         if (frame->tf_eflags & PSL_NT)
 950                 printf("nested task, ");
 951         if (frame->tf_eflags & PSL_RF)
 952                 printf("resume, ");
 953         if (frame->tf_eflags & PSL_VM)
 954                 printf("vm86, ");
 955         printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
 956         printf("current process         = ");
 957         if (curproc) {
 958                 printf("%lu (%s)\n", (u_long)curproc->p_pid, curthread->td_name);
 959         } else {
 960                 printf("Idle\n");
 961         }
 962
 963 #ifdef KDB
 964         if (debugger_on_panic || kdb_active) {
 965                 frame->tf_err = eva;    /* smuggle fault address to ddb */
 966                 if (kdb_trap(type, 0, frame)) {
 967                         frame->tf_err = code;   /* restore error code */
 968                         return;
 969                 }
 970                 frame->tf_err = code;           /* restore error code */
 971         }
 972 #endif
 973         printf("trap number             = %d\n", type);
 974         if (type <= MAX_TRAP_MSG)
 975                 panic("%s", trap_msg[type]);
 976         else
 977                 panic("unknown/reserved trap");
 978 }
 979
 980 /*
 981  * Double fault handler. Called when a fault occurs while writing
 982  * a frame for a trap/exception onto the stack. This usually occurs
 983  * when the stack overflows (such is the case with infinite recursion,
 984  * for example).
 985  *
 986  * XXX Note that the current PTD gets replaced by IdlePTD when the
 987  * task switch occurs. This means that the stack that was active at
 988  * the time of the double fault is not available at <kstack> unless
 989  * the machine was idle when the double fault occurred. The downside
 990  * of this is that "trace <ebp>" in ddb won't work.
 991  */
 992 void
 993 dblfault_handler()
 994 {
 995 #ifdef KDTRACE_HOOKS
 996         if (dtrace_doubletrap_func != NULL)
 997                 (*dtrace_doubletrap_func)();
 998 #endif
 999         printf("\nFatal double fault:\n");
1000         printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip));
1001         printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp));
1002         printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp));
1003 #ifdef SMP
1004         /* two separate prints in case of a trap on an unmapped page */
1005         printf("cpuid = %d; ", PCPU_GET(cpuid));
1006         printf("apic id = %02x\n", PCPU_GET(apic_id));
1007 #endif
1008         panic("double fault");
1009 }
1010
1011 int
1012 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
1013 {
1014         struct proc *p;
1015         struct trapframe *frame;
1016         caddr_t params;
1017         int error;
1018
1019         p = td->td_proc;
1020         frame = td->td_frame;
1021
1022         params = (caddr_t)frame->tf_esp + sizeof(int);
1023         sa->code = frame->tf_eax;
1024
1025         /*
1026          * Need to check if this is a 32 bit or 64 bit syscall.
1027          */
1028         if (sa->code == SYS_syscall) {
1029                 /*
1030                  * Code is first argument, followed by actual args.
1031                  */
1032                 sa->code = fuword(params);
1033                 params += sizeof(int);
1034         } else if (sa->code == SYS___syscall) {
1035                 /*
1036                  * Like syscall, but code is a quad, so as to maintain
1037                  * quad alignment for the rest of the arguments.
1038                  */
1039                 sa->code = fuword(params);
1040                 params += sizeof(quad_t);
1041         }
1042
1043         if (p->p_sysent->sv_mask)
1044                 sa->code &= p->p_sysent->sv_mask;
1045         if (sa->code >= p->p_sysent->sv_size)
1046                 sa->callp = &p->p_sysent->sv_table[0];
1047         else
1048                 sa->callp = &p->p_sysent->sv_table[sa->code];
1049         sa->narg = sa->callp->sy_narg;
1050
1051         if (params != NULL && sa->narg != 0)
1052                 error = copyin(params, (caddr_t)sa->args,
1053                     (u_int)(sa->narg * sizeof(int)));
1054         else
1055                 error = 0;
1056
1057         if (error == 0) {
1058                 td->td_retval[0] = 0;
1059                 td->td_retval[1] = frame->tf_edx;
1060         }
1061
1062         return (error);
1063 }
1064
1065 #include "../../kern/subr_syscall.c"
1066
1067 /*
1068  *      syscall -       system call request C handler
1069  *
1070  *      A system call is essentially treated as a trap.
1071  */
1072 void
1073 syscall(struct trapframe *frame)
1074 {
1075         struct thread *td;
1076         struct syscall_args sa;
1077         register_t orig_tf_eflags;
1078         int error;
1079         ksiginfo_t ksi;
1080
1081 #ifdef DIAGNOSTIC
1082         if (ISPL(frame->tf_cs) != SEL_UPL) {
1083                 panic("syscall");
1084                 /* NOT REACHED */
1085         }
1086 #endif
1087         orig_tf_eflags = frame->tf_eflags;
1088
1089         td = curthread;
1090         td->td_frame = frame;
1091
1092         error = syscallenter(td, &sa);
1093
1094         /*
1095          * Traced syscall.
1096          */
1097         if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1098                 frame->tf_eflags &= ~PSL_T;
1099                 ksiginfo_init_trap(&ksi);
1100                 ksi.ksi_signo = SIGTRAP;
1101                 ksi.ksi_code = TRAP_TRACE;
1102                 ksi.ksi_addr = (void *)frame->tf_eip;
1103                 trapsignal(td, &ksi);
1104         }
1105
1106         KASSERT(PCB_USER_FPU(td->td_pcb),
1107             ("System call %s returning with kernel FPU ctx leaked",
1108              syscallname(td->td_proc, sa.code)));
1109         KASSERT(td->td_pcb->pcb_save == &td->td_pcb->pcb_user_save,
1110             ("System call %s returning with mangled pcb_save",
1111              syscallname(td->td_proc, sa.code)));
1112
1113         syscallret(td, error, &sa);
1114 }