sys/i386/i386/trap.c

   1 /*-
   2  * Copyright (C) 1994, David Greenman
   3  * Copyright (c) 1990, 1993
   4  *      The Regents of the University of California.  All rights reserved.
   5  *
   6  * This code is derived from software contributed to Berkeley by
   7  * the University of Utah, and William Jolitz.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in the
  16  *    documentation and/or other materials provided with the distribution.
  17  * 3. All advertising materials mentioning features or use of this software
  18  *    must display the following acknowledgement:
  19  *      This product includes software developed by the University of
  20  *      California, Berkeley and its contributors.
  21  * 4. Neither the name of the University nor the names of its contributors
  22  *    may be used to endorse or promote products derived from this software
  23  *    without specific prior written permission.
  24  *
  25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  35  * SUCH DAMAGE.
  36  *
  37  *      from: @(#)trap.c        7.4 (Berkeley) 5/13/91
  38  */
  39
  40 #include <sys/cdefs.h>
  41 __FBSDID("$FreeBSD$");
  42
  43 /*
  44  * 386 Trap and System call handling
  45  */
  46
  47 #include "opt_clock.h"
  48 #include "opt_cpu.h"
  49 #include "opt_hwpmc_hooks.h"
  50 #include "opt_isa.h"
  51 #include "opt_kdb.h"
  52 #include "opt_npx.h"
  53 #include "opt_stack.h"
  54 #include "opt_trap.h"
  55
  56 #include <sys/param.h>
  57 #include <sys/bus.h>
  58 #include <sys/systm.h>
  59 #include <sys/proc.h>
  60 #include <sys/pioctl.h>
  61 #include <sys/ptrace.h>
  62 #include <sys/kdb.h>
  63 #include <sys/kernel.h>
  64 #include <sys/ktr.h>
  65 #include <sys/lock.h>
  66 #include <sys/mutex.h>
  67 #include <sys/resourcevar.h>
  68 #include <sys/signalvar.h>
  69 #include <sys/syscall.h>
  70 #include <sys/sysctl.h>
  71 #include <sys/sysent.h>
  72 #include <sys/uio.h>
  73 #include <sys/vmmeter.h>
  74 #ifdef HWPMC_HOOKS
  75 #include <sys/pmckern.h>
  76 PMC_SOFT_DEFINE( , , page_fault, all);
  77 PMC_SOFT_DEFINE( , , page_fault, read);
  78 PMC_SOFT_DEFINE( , , page_fault, write);
  79 #endif
  80 #include <security/audit/audit.h>
  81
  82 #include <vm/vm.h>
  83 #include <vm/vm_param.h>
  84 #include <vm/pmap.h>
  85 #include <vm/vm_kern.h>
  86 #include <vm/vm_map.h>
  87 #include <vm/vm_page.h>
  88 #include <vm/vm_extern.h>
  89
  90 #include <machine/cpu.h>
  91 #include <machine/intr_machdep.h>
  92 #include <x86/mca.h>
  93 #include <machine/md_var.h>
  94 #include <machine/pcb.h>
  95 #ifdef SMP
  96 #include <machine/smp.h>
  97 #endif
  98 #include <machine/stack.h>
  99 #include <machine/tss.h>
 100 #include <machine/vm86.h>
 101
 102 #ifdef POWERFAIL_NMI
 103 #include <sys/syslog.h>
 104 #include <machine/clock.h>
 105 #endif
 106
 107 #ifdef KDTRACE_HOOKS
 108 #include <sys/dtrace_bsd.h>
 109 #endif
 110
 111 extern void trap(struct trapframe *frame);
 112 extern void syscall(struct trapframe *frame);
 113
 114 static int trap_pfault(struct trapframe *, int, vm_offset_t);
 115 static void trap_fatal(struct trapframe *, vm_offset_t);
 116 void dblfault_handler(void);
 117
 118 extern inthand_t IDTVEC(lcall_syscall);
 119
 120 #define MAX_TRAP_MSG            32
 121 static char *trap_msg[] = {
 122         "",                                     /*  0 unused */
 123         "privileged instruction fault",         /*  1 T_PRIVINFLT */
 124         "",                                     /*  2 unused */
 125         "breakpoint instruction fault",         /*  3 T_BPTFLT */
 126         "",                                     /*  4 unused */
 127         "",                                     /*  5 unused */
 128         "arithmetic trap",                      /*  6 T_ARITHTRAP */
 129         "",                                     /*  7 unused */
 130         "",                                     /*  8 unused */
 131         "general protection fault",             /*  9 T_PROTFLT */
 132         "trace trap",                           /* 10 T_TRCTRAP */
 133         "",                                     /* 11 unused */
 134         "page fault",                           /* 12 T_PAGEFLT */
 135         "",                                     /* 13 unused */
 136         "alignment fault",                      /* 14 T_ALIGNFLT */
 137         "",                                     /* 15 unused */
 138         "",                                     /* 16 unused */
 139         "",                                     /* 17 unused */
 140         "integer divide fault",                 /* 18 T_DIVIDE */
 141         "non-maskable interrupt trap",          /* 19 T_NMI */
 142         "overflow trap",                        /* 20 T_OFLOW */
 143         "FPU bounds check fault",               /* 21 T_BOUND */
 144         "FPU device not available",             /* 22 T_DNA */
 145         "double fault",                         /* 23 T_DOUBLEFLT */
 146         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
 147         "invalid TSS fault",                    /* 25 T_TSSFLT */
 148         "segment not present fault",            /* 26 T_SEGNPFLT */
 149         "stack fault",                          /* 27 T_STKFLT */
 150         "machine check trap",                   /* 28 T_MCHK */
 151         "SIMD floating-point exception",        /* 29 T_XMMFLT */
 152         "reserved (unknown) fault",             /* 30 T_RESERVED */
 153         "",                                     /* 31 unused (reserved) */
 154         "DTrace pid return trap",               /* 32 T_DTRACE_RET */
 155 };
 156
 157 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 158 int has_f00f_bug = 0;           /* Initialized so that it can be patched. */
 159 #endif
 160
 161 static int prot_fault_translation = 0;
 162 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
 163         &prot_fault_translation, 0, "Select signal to deliver on protection fault");
 164 static int uprintf_signal;
 165 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW,
 166     &uprintf_signal, 0,
 167     "Print debugging information on trap signal to ctty");
 168
 169 /*
 170  * Exception, fault, and trap interface to the FreeBSD kernel.
 171  * This common code is called from assembly language IDT gate entry
 172  * routines that prepare a suitable stack frame, and restore this
 173  * frame after the exception has been processed.
 174  */
 175
 176 void
 177 trap(struct trapframe *frame)
 178 {
 179 #ifdef KDTRACE_HOOKS
 180         struct reg regs;
 181 #endif
 182         struct thread *td = curthread;
 183         struct proc *p = td->td_proc;
 184 #ifdef KDB
 185         register_t dr6;
 186 #endif
 187         int i = 0, ucode = 0;
 188         u_int type;
 189         register_t addr = 0;
 190         vm_offset_t eva;
 191         ksiginfo_t ksi;
 192 #ifdef POWERFAIL_NMI
 193         static int lastalert = 0;
 194 #endif
 195
 196         PCPU_INC(cnt.v_trap);
 197         type = frame->tf_trapno;
 198
 199 #ifdef SMP
 200         /* Handler for NMI IPIs used for stopping CPUs. */
 201         if (type == T_NMI) {
 202                  if (ipi_nmi_handler() == 0)
 203                            goto out;
 204         }
 205 #endif /* SMP */
 206
 207 #ifdef KDB
 208         if (kdb_active) {
 209                 kdb_reenter();
 210                 goto out;
 211         }
 212 #endif
 213
 214         if (type == T_RESERVED) {
 215                 trap_fatal(frame, 0);
 216                 goto out;
 217         }
 218
 219         if (type == T_NMI) {
 220 #ifdef HWPMC_HOOKS
 221                 /*
 222                  * CPU PMCs interrupt using an NMI so we check for that first.
 223                  * If the HWPMC module is active, 'pmc_hook' will point to
 224                  * the function to be called.  A non-zero return value from the
 225                  * hook means that the NMI was consumed by it and that we can
 226                  * return immediately.
 227                  */
 228                 if (pmc_intr != NULL &&
 229                     (*pmc_intr)(PCPU_GET(cpuid), frame) != 0)
 230                         goto out;
 231 #endif
 232
 233 #ifdef STACK
 234                 if (stack_nmi_handler(frame) != 0)
 235                         goto out;
 236 #endif
 237         }
 238
 239         if (type == T_MCHK) {
 240                 mca_intr();
 241                 goto out;
 242         }
 243
 244 #ifdef KDTRACE_HOOKS
 245         /*
 246          * A trap can occur while DTrace executes a probe. Before
 247          * executing the probe, DTrace blocks re-scheduling and sets
 248          * a flag in its per-cpu flags to indicate that it doesn't
 249          * want to fault. On returning from the probe, the no-fault
 250          * flag is cleared and finally re-scheduling is enabled.
 251          */
 252         if ((type == T_PROTFLT || type == T_PAGEFLT) &&
 253             dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
 254                 goto out;
 255 #endif
 256
 257         if ((frame->tf_eflags & PSL_I) == 0) {
 258                 /*
 259                  * Buggy application or kernel code has disabled
 260                  * interrupts and then trapped.  Enabling interrupts
 261                  * now is wrong, but it is better than running with
 262                  * interrupts disabled until they are accidentally
 263                  * enabled later.
 264                  */
 265                 if (TRAPF_USERMODE(frame) &&
 266                     (curpcb->pcb_flags & PCB_VM86CALL) == 0)
 267                         uprintf(
 268                             "pid %ld (%s): trap %d with interrupts disabled\n",
 269                             (long)curproc->p_pid, curthread->td_name, type);
 270                 else if (type != T_NMI && type != T_BPTFLT &&
 271                     type != T_TRCTRAP &&
 272                     frame->tf_eip != (int)cpu_switch_load_gs) {
 273                         /*
 274                          * XXX not quite right, since this may be for a
 275                          * multiple fault in user mode.
 276                          */
 277                         printf("kernel trap %d with interrupts disabled\n",
 278                             type);
 279                         /*
 280                          * Page faults need interrupts disabled until later,
 281                          * and we shouldn't enable interrupts while holding
 282                          * a spin lock.
 283                          */
 284                         if (type != T_PAGEFLT &&
 285                             td->td_md.md_spinlock_count == 0)
 286                                 enable_intr();
 287                 }
 288         }
 289         eva = 0;
 290         if (type == T_PAGEFLT) {
 291                 /*
 292                  * For some Cyrix CPUs, %cr2 is clobbered by
 293                  * interrupts.  This problem is worked around by using
 294                  * an interrupt gate for the pagefault handler.  We
 295                  * are finally ready to read %cr2 and conditionally
 296                  * reenable interrupts.  If we hold a spin lock, then
 297                  * we must not reenable interrupts.  This might be a
 298                  * spurious page fault.
 299                  */
 300                 eva = rcr2();
 301                 if (td->td_md.md_spinlock_count == 0)
 302                         enable_intr();
 303         }
 304
 305         if (TRAPF_USERMODE(frame) && (curpcb->pcb_flags & PCB_VM86CALL) == 0) {
 306                 /* user trap */
 307
 308                 td->td_pticks = 0;
 309                 td->td_frame = frame;
 310                 addr = frame->tf_eip;
 311                 if (td->td_cowgen != p->p_cowgen)
 312                         thread_cow_update(td);
 313
 314                 switch (type) {
 315                 case T_PRIVINFLT:       /* privileged instruction fault */
 316                         i = SIGILL;
 317                         ucode = ILL_PRVOPC;
 318                         break;
 319
 320                 case T_BPTFLT:          /* bpt instruction fault */
 321                 case T_TRCTRAP:         /* trace trap */
 322                         enable_intr();
 323 #ifdef KDTRACE_HOOKS
 324                         if (type == T_BPTFLT) {
 325                                 fill_frame_regs(frame, &regs);
 326                                 if (dtrace_pid_probe_ptr != NULL &&
 327                                     dtrace_pid_probe_ptr(&regs) == 0)
 328                                         goto out;
 329                         }
 330 #endif
 331 user_trctrap_out:
 332                         frame->tf_eflags &= ~PSL_T;
 333                         i = SIGTRAP;
 334                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 335                         break;
 336
 337                 case T_ARITHTRAP:       /* arithmetic trap */
 338 #ifdef DEV_NPX
 339                         ucode = npxtrap_x87();
 340                         if (ucode == -1)
 341                                 goto userout;
 342 #else
 343                         ucode = 0;
 344 #endif
 345                         i = SIGFPE;
 346                         break;
 347
 348                         /*
 349                          * The following two traps can happen in
 350                          * vm86 mode, and, if so, we want to handle
 351                          * them specially.
 352                          */
 353                 case T_PROTFLT:         /* general protection fault */
 354                 case T_STKFLT:          /* stack fault */
 355                         if (frame->tf_eflags & PSL_VM) {
 356                                 i = vm86_emulate((struct vm86frame *)frame);
 357                                 if (i == SIGTRAP) {
 358                                         type = T_TRCTRAP;
 359                                         load_dr6(rdr6() | 0x4000);
 360                                         goto user_trctrap_out;
 361                                 }
 362                                 if (i == 0)
 363                                         goto user;
 364                                 break;
 365                         }
 366                         i = SIGBUS;
 367                         ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
 368                         break;
 369                 case T_SEGNPFLT:        /* segment not present fault */
 370                         i = SIGBUS;
 371                         ucode = BUS_ADRERR;
 372                         break;
 373                 case T_TSSFLT:          /* invalid TSS fault */
 374                         i = SIGBUS;
 375                         ucode = BUS_OBJERR;
 376                         break;
 377                 case T_ALIGNFLT:
 378                         i = SIGBUS;
 379                         ucode = BUS_ADRALN;
 380                         break;
 381                 case T_DOUBLEFLT:       /* double fault */
 382                 default:
 383                         i = SIGBUS;
 384                         ucode = BUS_OBJERR;
 385                         break;
 386
 387                 case T_PAGEFLT:         /* page fault */
 388
 389                         i = trap_pfault(frame, TRUE, eva);
 390 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 391                         if (i == -2) {
 392                                 /*
 393                                  * The f00f hack workaround has triggered, so
 394                                  * treat the fault as an illegal instruction
 395                                  * (T_PRIVINFLT) instead of a page fault.
 396                                  */
 397                                 type = frame->tf_trapno = T_PRIVINFLT;
 398
 399                                 /* Proceed as in that case. */
 400                                 ucode = ILL_PRVOPC;
 401                                 i = SIGILL;
 402                                 break;
 403                         }
 404 #endif
 405                         if (i == -1)
 406                                 goto userout;
 407                         if (i == 0)
 408                                 goto user;
 409
 410                         if (i == SIGSEGV)
 411                                 ucode = SEGV_MAPERR;
 412                         else {
 413                                 if (prot_fault_translation == 0) {
 414                                         /*
 415                                          * Autodetect.
 416                                          * This check also covers the images
 417                                          * without the ABI-tag ELF note.
 418                                          */
 419                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 420                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 421                                                 i = SIGSEGV;
 422                                                 ucode = SEGV_ACCERR;
 423                                         } else {
 424                                                 i = SIGBUS;
 425                                                 ucode = BUS_PAGE_FAULT;
 426                                         }
 427                                 } else if (prot_fault_translation == 1) {
 428                                         /*
 429                                          * Always compat mode.
 430                                          */
 431                                         i = SIGBUS;
 432                                         ucode = BUS_PAGE_FAULT;
 433                                 } else {
 434                                         /*
 435                                          * Always SIGSEGV mode.
 436                                          */
 437                                         i = SIGSEGV;
 438                                         ucode = SEGV_ACCERR;
 439                                 }
 440                         }
 441                         addr = eva;
 442                         break;
 443
 444                 case T_DIVIDE:          /* integer divide fault */
 445                         ucode = FPE_INTDIV;
 446                         i = SIGFPE;
 447                         break;
 448
 449 #ifdef DEV_ISA
 450                 case T_NMI:
 451 #ifdef POWERFAIL_NMI
 452 #ifndef TIMER_FREQ
 453 #  define TIMER_FREQ 1193182
 454 #endif
 455                         if (time_second - lastalert > 10) {
 456                                 log(LOG_WARNING, "NMI: power fail\n");
 457                                 sysbeep(880, hz);
 458                                 lastalert = time_second;
 459                         }
 460                         goto userout;
 461 #else /* !POWERFAIL_NMI */
 462                         nmi_handle_intr(type, frame);
 463                         break;
 464 #endif /* POWERFAIL_NMI */
 465 #endif /* DEV_ISA */
 466
 467                 case T_OFLOW:           /* integer overflow fault */
 468                         ucode = FPE_INTOVF;
 469                         i = SIGFPE;
 470                         break;
 471
 472                 case T_BOUND:           /* bounds check fault */
 473                         ucode = FPE_FLTSUB;
 474                         i = SIGFPE;
 475                         break;
 476
 477                 case T_DNA:
 478 #ifdef DEV_NPX
 479                         KASSERT(PCB_USER_FPU(td->td_pcb),
 480                             ("kernel FPU ctx has leaked"));
 481                         /* transparent fault (due to context switch "late") */
 482                         if (npxdna())
 483                                 goto userout;
 484 #endif
 485                         uprintf("pid %d killed due to lack of floating point\n",
 486                                 p->p_pid);
 487                         i = SIGKILL;
 488                         ucode = 0;
 489                         break;
 490
 491                 case T_FPOPFLT:         /* FPU operand fetch fault */
 492                         ucode = ILL_COPROC;
 493                         i = SIGILL;
 494                         break;
 495
 496                 case T_XMMFLT:          /* SIMD floating-point exception */
 497 #if defined(DEV_NPX) && !defined(CPU_DISABLE_SSE) && defined(I686_CPU)
 498                         ucode = npxtrap_sse();
 499                         if (ucode == -1)
 500                                 goto userout;
 501 #else
 502                         ucode = 0;
 503 #endif
 504                         i = SIGFPE;
 505                         break;
 506 #ifdef KDTRACE_HOOKS
 507                 case T_DTRACE_RET:
 508                         enable_intr();
 509                         fill_frame_regs(frame, &regs);
 510                         if (dtrace_return_probe_ptr != NULL &&
 511                             dtrace_return_probe_ptr(&regs) == 0)
 512                                 goto out;
 513                         break;
 514 #endif
 515                 }
 516         } else {
 517                 /* kernel trap */
 518
 519                 KASSERT(cold || td->td_ucred != NULL,
 520                     ("kernel trap doesn't have ucred"));
 521                 switch (type) {
 522                 case T_PAGEFLT:                 /* page fault */
 523                         (void) trap_pfault(frame, FALSE, eva);
 524                         goto out;
 525
 526                 case T_DNA:
 527 #ifdef DEV_NPX
 528                         if (PCB_USER_FPU(td->td_pcb))
 529                                 panic("Unregistered use of FPU in kernel");
 530                         if (npxdna())
 531                                 goto out;
 532 #endif
 533                         break;
 534
 535                 case T_ARITHTRAP:       /* arithmetic trap */
 536                 case T_XMMFLT:          /* SIMD floating-point exception */
 537                 case T_FPOPFLT:         /* FPU operand fetch fault */
 538                         /*
 539                          * XXXKIB for now disable any FPU traps in kernel
 540                          * handler registration seems to be overkill
 541                          */
 542                         trap_fatal(frame, 0);
 543                         goto out;
 544
 545                         /*
 546                          * The following two traps can happen in
 547                          * vm86 mode, and, if so, we want to handle
 548                          * them specially.
 549                          */
 550                 case T_PROTFLT:         /* general protection fault */
 551                 case T_STKFLT:          /* stack fault */
 552                         if (frame->tf_eflags & PSL_VM) {
 553                                 i = vm86_emulate((struct vm86frame *)frame);
 554                                 if (i == SIGTRAP) {
 555                                         type = T_TRCTRAP;
 556                                         load_dr6(rdr6() | 0x4000);
 557                                         goto kernel_trctrap;
 558                                 }
 559                                 if (i != 0)
 560                                         /*
 561                                          * returns to original process
 562                                          */
 563                                         vm86_trap((struct vm86frame *)frame);
 564                                 goto out;
 565                         }
 566                         if (type == T_STKFLT)
 567                                 break;
 568
 569                         /* FALL THROUGH */
 570
 571                 case T_SEGNPFLT:        /* segment not present fault */
 572                         if (curpcb->pcb_flags & PCB_VM86CALL)
 573                                 break;
 574
 575                         /*
 576                          * Invalid %fs's and %gs's can be created using
 577                          * procfs or PT_SETREGS or by invalidating the
 578                          * underlying LDT entry.  This causes a fault
 579                          * in kernel mode when the kernel attempts to
 580                          * switch contexts.  Lose the bad context
 581                          * (XXX) so that we can continue, and generate
 582                          * a signal.
 583                          */
 584                         if (frame->tf_eip == (int)cpu_switch_load_gs) {
 585                                 curpcb->pcb_gs = 0;
 586 #if 0
 587                                 PROC_LOCK(p);
 588                                 kern_psignal(p, SIGBUS);
 589                                 PROC_UNLOCK(p);
 590 #endif
 591                                 goto out;
 592                         }
 593
 594                         if (td->td_intr_nesting_level != 0)
 595                                 break;
 596
 597                         /*
 598                          * Invalid segment selectors and out of bounds
 599                          * %eip's and %esp's can be set up in user mode.
 600                          * This causes a fault in kernel mode when the
 601                          * kernel tries to return to user mode.  We want
 602                          * to get this fault so that we can fix the
 603                          * problem here and not have to check all the
 604                          * selectors and pointers when the user changes
 605                          * them.
 606                          */
 607                         if (frame->tf_eip == (int)doreti_iret) {
 608                                 frame->tf_eip = (int)doreti_iret_fault;
 609                                 goto out;
 610                         }
 611                         if (frame->tf_eip == (int)doreti_popl_ds) {
 612                                 frame->tf_eip = (int)doreti_popl_ds_fault;
 613                                 goto out;
 614                         }
 615                         if (frame->tf_eip == (int)doreti_popl_es) {
 616                                 frame->tf_eip = (int)doreti_popl_es_fault;
 617                                 goto out;
 618                         }
 619                         if (frame->tf_eip == (int)doreti_popl_fs) {
 620                                 frame->tf_eip = (int)doreti_popl_fs_fault;
 621                                 goto out;
 622                         }
 623                         if (curpcb->pcb_onfault != NULL) {
 624                                 frame->tf_eip =
 625                                     (int)curpcb->pcb_onfault;
 626                                 goto out;
 627                         }
 628                         break;
 629
 630                 case T_TSSFLT:
 631                         /*
 632                          * PSL_NT can be set in user mode and isn't cleared
 633                          * automatically when the kernel is entered.  This
 634                          * causes a TSS fault when the kernel attempts to
 635                          * `iret' because the TSS link is uninitialized.  We
 636                          * want to get this fault so that we can fix the
 637                          * problem here and not every time the kernel is
 638                          * entered.
 639                          */
 640                         if (frame->tf_eflags & PSL_NT) {
 641                                 frame->tf_eflags &= ~PSL_NT;
 642                                 goto out;
 643                         }
 644                         break;
 645
 646                 case T_TRCTRAP:  /* trace trap */
 647 kernel_trctrap:
 648                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall)) {
 649                                 /*
 650                                  * We've just entered system mode via the
 651                                  * syscall lcall.  Continue single stepping
 652                                  * silently until the syscall handler has
 653                                  * saved the flags.
 654                                  */
 655                                 goto out;
 656                         }
 657                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall) + 1) {
 658                                 /*
 659                                  * The syscall handler has now saved the
 660                                  * flags.  Stop single stepping it.
 661                                  */
 662                                 frame->tf_eflags &= ~PSL_T;
 663                                 goto out;
 664                         }
 665                         /*
 666                          * Ignore debug register trace traps due to
 667                          * accesses in the user's address space, which
 668                          * can happen under several conditions such as
 669                          * if a user sets a watchpoint on a buffer and
 670                          * then passes that buffer to a system call.
 671                          * We still want to get TRCTRAPS for addresses
 672                          * in kernel space because that is useful when
 673                          * debugging the kernel.
 674                          */
 675                         if (user_dbreg_trap() &&
 676                            !(curpcb->pcb_flags & PCB_VM86CALL)) {
 677                                 /*
 678                                  * Reset breakpoint bits because the
 679                                  * processor doesn't
 680                                  */
 681                                 load_dr6(rdr6() & ~0xf);
 682                                 goto out;
 683                         }
 684                         /*
 685                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 686                          */
 687                 case T_BPTFLT:
 688                         /*
 689                          * If KDB is enabled, let it handle the debugger trap.
 690                          * Otherwise, debugger traps "can't happen".
 691                          */
 692 #ifdef KDB
 693                         /* XXX %dr6 is not quite reentrant. */
 694                         dr6 = rdr6();
 695                         load_dr6(dr6 & ~0x4000);
 696                         if (kdb_trap(type, dr6, frame))
 697                                 goto out;
 698 #endif
 699                         break;
 700
 701 #ifdef DEV_ISA
 702                 case T_NMI:
 703 #ifdef POWERFAIL_NMI
 704                         if (time_second - lastalert > 10) {
 705                                 log(LOG_WARNING, "NMI: power fail\n");
 706                                 sysbeep(880, hz);
 707                                 lastalert = time_second;
 708                         }
 709                         goto out;
 710 #else /* !POWERFAIL_NMI */
 711                         nmi_handle_intr(type, frame);
 712                         goto out;
 713 #endif /* POWERFAIL_NMI */
 714 #endif /* DEV_ISA */
 715                 }
 716
 717                 trap_fatal(frame, eva);
 718                 goto out;
 719         }
 720
 721         /* Translate fault for emulators (e.g. Linux) */
 722         if (*p->p_sysent->sv_transtrap)
 723                 i = (*p->p_sysent->sv_transtrap)(i, type);
 724
 725         ksiginfo_init_trap(&ksi);
 726         ksi.ksi_signo = i;
 727         ksi.ksi_code = ucode;
 728         ksi.ksi_addr = (void *)addr;
 729         ksi.ksi_trapno = type;
 730         if (uprintf_signal) {
 731                 uprintf("pid %d comm %s: signal %d err %x code %d type %d "
 732                     "addr 0x%x esp 0x%08x eip 0x%08x "
 733                     "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
 734                     p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr,
 735                     frame->tf_esp, frame->tf_eip,
 736                     fubyte((void *)(frame->tf_eip + 0)),
 737                     fubyte((void *)(frame->tf_eip + 1)),
 738                     fubyte((void *)(frame->tf_eip + 2)),
 739                     fubyte((void *)(frame->tf_eip + 3)),
 740                     fubyte((void *)(frame->tf_eip + 4)),
 741                     fubyte((void *)(frame->tf_eip + 5)),
 742                     fubyte((void *)(frame->tf_eip + 6)),
 743                     fubyte((void *)(frame->tf_eip + 7)));
 744         }
 745         KASSERT((read_eflags() & PSL_I) != 0, ("interrupts disabled"));
 746         trapsignal(td, &ksi);
 747
 748 #ifdef DEBUG
 749         if (type <= MAX_TRAP_MSG) {
 750                 uprintf("fatal process exception: %s",
 751                         trap_msg[type]);
 752                 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
 753                         uprintf(", fault VA = 0x%lx", (u_long)eva);
 754                 uprintf("\n");
 755         }
 756 #endif
 757
 758 user:
 759         userret(td, frame);
 760         KASSERT(PCB_USER_FPU(td->td_pcb),
 761             ("Return from trap with kernel FPU ctx leaked"));
 762 userout:
 763 out:
 764         return;
 765 }
 766
 767 static int
 768 trap_pfault(frame, usermode, eva)
 769         struct trapframe *frame;
 770         int usermode;
 771         vm_offset_t eva;
 772 {
 773         vm_offset_t va;
 774         vm_map_t map;
 775         int rv = 0;
 776         vm_prot_t ftype;
 777         struct thread *td = curthread;
 778         struct proc *p = td->td_proc;
 779
 780         if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
 781                 /*
 782                  * Due to both processor errata and lazy TLB invalidation when
 783                  * access restrictions are removed from virtual pages, memory
 784                  * accesses that are allowed by the physical mapping layer may
 785                  * nonetheless cause one spurious page fault per virtual page.
 786                  * When the thread is executing a "no faulting" section that
 787                  * is bracketed by vm_fault_{disable,enable}_pagefaults(),
 788                  * every page fault is treated as a spurious page fault,
 789                  * unless it accesses the same virtual address as the most
 790                  * recent page fault within the same "no faulting" section.
 791                  */
 792                 if (td->td_md.md_spurflt_addr != eva ||
 793                     (td->td_pflags & TDP_RESETSPUR) != 0) {
 794                         /*
 795                          * Do nothing to the TLB.  A stale TLB entry is
 796                          * flushed automatically by a page fault.
 797                          */
 798                         td->td_md.md_spurflt_addr = eva;
 799                         td->td_pflags &= ~TDP_RESETSPUR;
 800                         return (0);
 801                 }
 802         } else {
 803                 /*
 804                  * If we get a page fault while in a critical section, then
 805                  * it is most likely a fatal kernel page fault.  The kernel
 806                  * is already going to panic trying to get a sleep lock to
 807                  * do the VM lookup, so just consider it a fatal trap so the
 808                  * kernel can print out a useful trap message and even get
 809                  * to the debugger.
 810                  *
 811                  * If we get a page fault while holding a non-sleepable
 812                  * lock, then it is most likely a fatal kernel page fault.
 813                  * If WITNESS is enabled, then it's going to whine about
 814                  * bogus LORs with various VM locks, so just skip to the
 815                  * fatal trap handling directly.
 816                  */
 817                 if (td->td_critnest != 0 ||
 818                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 819                     "Kernel page fault") != 0) {
 820                         trap_fatal(frame, eva);
 821                         return (-1);
 822                 }
 823         }
 824         va = trunc_page(eva);
 825         if (va >= KERNBASE) {
 826                 /*
 827                  * Don't allow user-mode faults in kernel address space.
 828                  * An exception:  if the faulting address is the invalid
 829                  * instruction entry in the IDT, then the Intel Pentium
 830                  * F00F bug workaround was triggered, and we need to
 831                  * treat it is as an illegal instruction, and not a page
 832                  * fault.
 833                  */
 834 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 835                 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
 836                         return (-2);
 837 #endif
 838                 if (usermode)
 839                         goto nogo;
 840
 841                 map = kernel_map;
 842         } else {
 843                 map = &p->p_vmspace->vm_map;
 844
 845                 /*
 846                  * When accessing a user-space address, kernel must be
 847                  * ready to accept the page fault, and provide a
 848                  * handling routine.  Since accessing the address
 849                  * without the handler is a bug, do not try to handle
 850                  * it normally, and panic immediately.
 851                  */
 852                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 853                     curpcb->pcb_onfault == NULL)) {
 854                         trap_fatal(frame, eva);
 855                         return (-1);
 856                 }
 857         }
 858
 859         /*
 860          * If the trap was caused by errant bits in the PTE then panic.
 861          */
 862         if (frame->tf_err & PGEX_RSV) {
 863                 trap_fatal(frame, eva);
 864                 return (-1);
 865         }
 866
 867         /*
 868          * PGEX_I is defined only if the execute disable bit capability is
 869          * supported and enabled.
 870          */
 871         if (frame->tf_err & PGEX_W)
 872                 ftype = VM_PROT_WRITE;
 873 #if defined(PAE) || defined(PAE_TABLES)
 874         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 875                 ftype = VM_PROT_EXECUTE;
 876 #endif
 877         else
 878                 ftype = VM_PROT_READ;
 879
 880         /* Fault in the page. */
 881         rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 882         if (rv == KERN_SUCCESS) {
 883 #ifdef HWPMC_HOOKS
 884                 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
 885                         PMC_SOFT_CALL_TF( , , page_fault, all, frame);
 886                         if (ftype == VM_PROT_READ)
 887                                 PMC_SOFT_CALL_TF( , , page_fault, read,
 888                                     frame);
 889                         else
 890                                 PMC_SOFT_CALL_TF( , , page_fault, write,
 891                                     frame);
 892                 }
 893 #endif
 894                 return (0);
 895         }
 896 nogo:
 897         if (!usermode) {
 898                 if (td->td_intr_nesting_level == 0 &&
 899                     curpcb->pcb_onfault != NULL) {
 900                         frame->tf_eip = (int)curpcb->pcb_onfault;
 901                         return (0);
 902                 }
 903                 trap_fatal(frame, eva);
 904                 return (-1);
 905         }
 906         return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 907 }
 908
 909 static void
 910 trap_fatal(frame, eva)
 911         struct trapframe *frame;
 912         vm_offset_t eva;
 913 {
 914         int code, ss, esp;
 915         u_int type;
 916         struct soft_segment_descriptor softseg;
 917         char *msg;
 918
 919         code = frame->tf_err;
 920         type = frame->tf_trapno;
 921         sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
 922
 923         if (type <= MAX_TRAP_MSG)
 924                 msg = trap_msg[type];
 925         else
 926                 msg = "UNKNOWN";
 927         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 928             frame->tf_eflags & PSL_VM ? "vm86" :
 929             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
 930 #ifdef SMP
 931         /* two separate prints in case of a trap on an unmapped page */
 932         printf("cpuid = %d; ", PCPU_GET(cpuid));
 933         printf("apic id = %02x\n", PCPU_GET(apic_id));
 934 #endif
 935         if (type == T_PAGEFLT) {
 936                 printf("fault virtual address   = 0x%x\n", eva);
 937                 printf("fault code              = %s %s%s, %s\n",
 938                         code & PGEX_U ? "user" : "supervisor",
 939                         code & PGEX_W ? "write" : "read",
 940 #if defined(PAE) || defined(PAE_TABLES)
 941                         pg_nx != 0 ?
 942                         (code & PGEX_I ? " instruction" : " data") :
 943 #endif
 944                         "",
 945                         code & PGEX_RSV ? "reserved bits in PTE" :
 946                         code & PGEX_P ? "protection violation" : "page not present");
 947         }
 948         printf("instruction pointer     = 0x%x:0x%x\n",
 949                frame->tf_cs & 0xffff, frame->tf_eip);
 950         if (TF_HAS_STACKREGS(frame)) {
 951                 ss = frame->tf_ss & 0xffff;
 952                 esp = frame->tf_esp;
 953         } else {
 954                 ss = GSEL(GDATA_SEL, SEL_KPL);
 955                 esp = (int)&frame->tf_esp;
 956         }
 957         printf("stack pointer           = 0x%x:0x%x\n", ss, esp);
 958         printf("frame pointer           = 0x%x:0x%x\n", ss, frame->tf_ebp);
 959         printf("code segment            = base 0x%x, limit 0x%x, type 0x%x\n",
 960                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 961         printf("                        = DPL %d, pres %d, def32 %d, gran %d\n",
 962                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
 963                softseg.ssd_gran);
 964         printf("processor eflags        = ");
 965         if (frame->tf_eflags & PSL_T)
 966                 printf("trace trap, ");
 967         if (frame->tf_eflags & PSL_I)
 968                 printf("interrupt enabled, ");
 969         if (frame->tf_eflags & PSL_NT)
 970                 printf("nested task, ");
 971         if (frame->tf_eflags & PSL_RF)
 972                 printf("resume, ");
 973         if (frame->tf_eflags & PSL_VM)
 974                 printf("vm86, ");
 975         printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
 976         printf("current process         = %d (%s)\n",
 977             curproc->p_pid, curthread->td_name);
 978
 979 #ifdef KDB
 980         if (debugger_on_panic || kdb_active) {
 981                 frame->tf_err = eva;    /* smuggle fault address to ddb */
 982                 if (kdb_trap(type, 0, frame)) {
 983                         frame->tf_err = code;   /* restore error code */
 984                         return;
 985                 }
 986                 frame->tf_err = code;           /* restore error code */
 987         }
 988 #endif
 989         printf("trap number             = %d\n", type);
 990         if (type <= MAX_TRAP_MSG)
 991                 panic("%s", trap_msg[type]);
 992         else
 993                 panic("unknown/reserved trap");
 994 }
 995
 996 /*
 997  * Double fault handler. Called when a fault occurs while writing
 998  * a frame for a trap/exception onto the stack. This usually occurs
 999  * when the stack overflows (such is the case with infinite recursion,
1000  * for example).
1001  *
1002  * XXX Note that the current PTD gets replaced by IdlePTD when the
1003  * task switch occurs. This means that the stack that was active at
1004  * the time of the double fault is not available at <kstack> unless
1005  * the machine was idle when the double fault occurred. The downside
1006  * of this is that "trace <ebp>" in ddb won't work.
1007  */
1008 void
1009 dblfault_handler()
1010 {
1011 #ifdef KDTRACE_HOOKS
1012         if (dtrace_doubletrap_func != NULL)
1013                 (*dtrace_doubletrap_func)();
1014 #endif
1015         printf("\nFatal double fault:\n");
1016         printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip));
1017         printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp));
1018         printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp));
1019 #ifdef SMP
1020         /* two separate prints in case of a trap on an unmapped page */
1021         printf("cpuid = %d; ", PCPU_GET(cpuid));
1022         printf("apic id = %02x\n", PCPU_GET(apic_id));
1023 #endif
1024         panic("double fault");
1025 }
1026
1027 int
1028 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
1029 {
1030         struct proc *p;
1031         struct trapframe *frame;
1032         caddr_t params;
1033         long tmp;
1034         int error;
1035
1036         p = td->td_proc;
1037         frame = td->td_frame;
1038
1039         params = (caddr_t)frame->tf_esp + sizeof(int);
1040         sa->code = frame->tf_eax;
1041
1042         /*
1043          * Need to check if this is a 32 bit or 64 bit syscall.
1044          */
1045         if (sa->code == SYS_syscall) {
1046                 /*
1047                  * Code is first argument, followed by actual args.
1048                  */
1049                 error = fueword(params, &tmp);
1050                 if (error == -1)
1051                         return (EFAULT);
1052                 sa->code = tmp;
1053                 params += sizeof(int);
1054         } else if (sa->code == SYS___syscall) {
1055                 /*
1056                  * Like syscall, but code is a quad, so as to maintain
1057                  * quad alignment for the rest of the arguments.
1058                  */
1059                 error = fueword(params, &tmp);
1060                 if (error == -1)
1061                         return (EFAULT);
1062                 sa->code = tmp;
1063                 params += sizeof(quad_t);
1064         }
1065
1066         if (p->p_sysent->sv_mask)
1067                 sa->code &= p->p_sysent->sv_mask;
1068         if (sa->code >= p->p_sysent->sv_size)
1069                 sa->callp = &p->p_sysent->sv_table[0];
1070         else
1071                 sa->callp = &p->p_sysent->sv_table[sa->code];
1072         sa->narg = sa->callp->sy_narg;
1073
1074         if (params != NULL && sa->narg != 0)
1075                 error = copyin(params, (caddr_t)sa->args,
1076                     (u_int)(sa->narg * sizeof(int)));
1077         else
1078                 error = 0;
1079
1080         if (error == 0) {
1081                 td->td_retval[0] = 0;
1082                 td->td_retval[1] = frame->tf_edx;
1083         }
1084
1085         return (error);
1086 }
1087
1088 #include "../../kern/subr_syscall.c"
1089
1090 /*
1091  * syscall - system call request C handler.  A system call is
1092  * essentially treated as a trap by reusing the frame layout.
1093  */
1094 void
1095 syscall(struct trapframe *frame)
1096 {
1097         struct thread *td;
1098         struct syscall_args sa;
1099         register_t orig_tf_eflags;
1100         int error;
1101         ksiginfo_t ksi;
1102
1103 #ifdef DIAGNOSTIC
1104         if (!(TRAPF_USERMODE(frame) &&
1105             (curpcb->pcb_flags & PCB_VM86CALL) == 0)) {
1106                 panic("syscall");
1107                 /* NOT REACHED */
1108         }
1109 #endif
1110         orig_tf_eflags = frame->tf_eflags;
1111
1112         td = curthread;
1113         td->td_frame = frame;
1114
1115         error = syscallenter(td, &sa);
1116
1117         /*
1118          * Traced syscall.
1119          */
1120         if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1121                 frame->tf_eflags &= ~PSL_T;
1122                 ksiginfo_init_trap(&ksi);
1123                 ksi.ksi_signo = SIGTRAP;
1124                 ksi.ksi_code = TRAP_TRACE;
1125                 ksi.ksi_addr = (void *)frame->tf_eip;
1126                 trapsignal(td, &ksi);
1127         }
1128
1129         KASSERT(PCB_USER_FPU(td->td_pcb),
1130             ("System call %s returning with kernel FPU ctx leaked",
1131              syscallname(td->td_proc, sa.code)));
1132         KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td),
1133             ("System call %s returning with mangled pcb_save",
1134              syscallname(td->td_proc, sa.code)));
1135
1136         syscallret(td, error, &sa);
1137 }