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_trap.h"
  54
  55 #include <sys/param.h>
  56 #include <sys/bus.h>
  57 #include <sys/systm.h>
  58 #include <sys/proc.h>
  59 #include <sys/pioctl.h>
  60 #include <sys/ptrace.h>
  61 #include <sys/kdb.h>
  62 #include <sys/kernel.h>
  63 #include <sys/ktr.h>
  64 #include <sys/lock.h>
  65 #include <sys/mutex.h>
  66 #include <sys/resourcevar.h>
  67 #include <sys/signalvar.h>
  68 #include <sys/syscall.h>
  69 #include <sys/sysctl.h>
  70 #include <sys/sysent.h>
  71 #include <sys/uio.h>
  72 #include <sys/vmmeter.h>
  73 #ifdef HWPMC_HOOKS
  74 #include <sys/pmckern.h>
  75 PMC_SOFT_DEFINE( , , page_fault, all);
  76 PMC_SOFT_DEFINE( , , page_fault, read);
  77 PMC_SOFT_DEFINE( , , page_fault, write);
  78 #endif
  79 #include <security/audit/audit.h>
  80
  81 #include <vm/vm.h>
  82 #include <vm/vm_param.h>
  83 #include <vm/pmap.h>
  84 #include <vm/vm_kern.h>
  85 #include <vm/vm_map.h>
  86 #include <vm/vm_page.h>
  87 #include <vm/vm_extern.h>
  88
  89 #include <machine/cpu.h>
  90 #include <machine/intr_machdep.h>
  91 #include <x86/mca.h>
  92 #include <machine/md_var.h>
  93 #include <machine/pcb.h>
  94 #ifdef SMP
  95 #include <machine/smp.h>
  96 #endif
  97 #include <machine/tss.h>
  98 #include <machine/vm86.h>
  99
 100 #ifdef POWERFAIL_NMI
 101 #include <sys/syslog.h>
 102 #include <machine/clock.h>
 103 #endif
 104
 105 #ifdef KDTRACE_HOOKS
 106 #include <sys/dtrace_bsd.h>
 107 #endif
 108
 109 extern void trap(struct trapframe *frame);
 110 extern void syscall(struct trapframe *frame);
 111
 112 static int trap_pfault(struct trapframe *, int, vm_offset_t);
 113 static void trap_fatal(struct trapframe *, vm_offset_t);
 114 void dblfault_handler(void);
 115
 116 extern inthand_t IDTVEC(lcall_syscall);
 117
 118 #define MAX_TRAP_MSG            32
 119 static char *trap_msg[] = {
 120         "",                                     /*  0 unused */
 121         "privileged instruction fault",         /*  1 T_PRIVINFLT */
 122         "",                                     /*  2 unused */
 123         "breakpoint instruction fault",         /*  3 T_BPTFLT */
 124         "",                                     /*  4 unused */
 125         "",                                     /*  5 unused */
 126         "arithmetic trap",                      /*  6 T_ARITHTRAP */
 127         "",                                     /*  7 unused */
 128         "",                                     /*  8 unused */
 129         "general protection fault",             /*  9 T_PROTFLT */
 130         "trace trap",                           /* 10 T_TRCTRAP */
 131         "",                                     /* 11 unused */
 132         "page fault",                           /* 12 T_PAGEFLT */
 133         "",                                     /* 13 unused */
 134         "alignment fault",                      /* 14 T_ALIGNFLT */
 135         "",                                     /* 15 unused */
 136         "",                                     /* 16 unused */
 137         "",                                     /* 17 unused */
 138         "integer divide fault",                 /* 18 T_DIVIDE */
 139         "non-maskable interrupt trap",          /* 19 T_NMI */
 140         "overflow trap",                        /* 20 T_OFLOW */
 141         "FPU bounds check fault",               /* 21 T_BOUND */
 142         "FPU device not available",             /* 22 T_DNA */
 143         "double fault",                         /* 23 T_DOUBLEFLT */
 144         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
 145         "invalid TSS fault",                    /* 25 T_TSSFLT */
 146         "segment not present fault",            /* 26 T_SEGNPFLT */
 147         "stack fault",                          /* 27 T_STKFLT */
 148         "machine check trap",                   /* 28 T_MCHK */
 149         "SIMD floating-point exception",        /* 29 T_XMMFLT */
 150         "reserved (unknown) fault",             /* 30 T_RESERVED */
 151         "",                                     /* 31 unused (reserved) */
 152         "DTrace pid return trap",               /* 32 T_DTRACE_RET */
 153 };
 154
 155 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 156 extern int has_f00f_bug;
 157 #endif
 158
 159 #ifdef KDB
 160 static int kdb_on_nmi = 1;
 161 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RWTUN,
 162         &kdb_on_nmi, 0, "Go to KDB on NMI");
 163 #endif
 164 static int panic_on_nmi = 1;
 165 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RWTUN,
 166         &panic_on_nmi, 0, "Panic on NMI");
 167 static int prot_fault_translation = 0;
 168 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
 169         &prot_fault_translation, 0, "Select signal to deliver on protection fault");
 170 static int uprintf_signal;
 171 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW,
 172     &uprintf_signal, 0,
 173     "Print debugging information on trap signal to ctty");
 174
 175 /*
 176  * Exception, fault, and trap interface to the FreeBSD kernel.
 177  * This common code is called from assembly language IDT gate entry
 178  * routines that prepare a suitable stack frame, and restore this
 179  * frame after the exception has been processed.
 180  */
 181
 182 void
 183 trap(struct trapframe *frame)
 184 {
 185 #ifdef KDTRACE_HOOKS
 186         struct reg regs;
 187 #endif
 188         struct thread *td = curthread;
 189         struct proc *p = td->td_proc;
 190         int i = 0, ucode = 0, code;
 191         u_int type;
 192         register_t addr = 0;
 193         vm_offset_t eva;
 194         ksiginfo_t ksi;
 195 #ifdef POWERFAIL_NMI
 196         static int lastalert = 0;
 197 #endif
 198
 199         PCPU_INC(cnt.v_trap);
 200         type = frame->tf_trapno;
 201
 202 #ifdef SMP
 203         /* Handler for NMI IPIs used for stopping CPUs. */
 204         if (type == T_NMI) {
 205                  if (ipi_nmi_handler() == 0)
 206                            goto out;
 207         }
 208 #endif /* SMP */
 209
 210 #ifdef KDB
 211         if (kdb_active) {
 212                 kdb_reenter();
 213                 goto out;
 214         }
 215 #endif
 216
 217         if (type == T_RESERVED) {
 218                 trap_fatal(frame, 0);
 219                 goto out;
 220         }
 221
 222 #ifdef  HWPMC_HOOKS
 223         /*
 224          * CPU PMCs interrupt using an NMI so we check for that first.
 225          * If the HWPMC module is active, 'pmc_hook' will point to
 226          * the function to be called.  A return value of '1' from the
 227          * hook means that the NMI was handled by it and that we can
 228          * return immediately.
 229          */
 230         if (type == T_NMI && pmc_intr &&
 231             (*pmc_intr)(PCPU_GET(cpuid), frame))
 232             goto out;
 233 #endif
 234
 235         if (type == T_MCHK) {
 236                 mca_intr();
 237                 goto out;
 238         }
 239
 240 #ifdef KDTRACE_HOOKS
 241         /*
 242          * A trap can occur while DTrace executes a probe. Before
 243          * executing the probe, DTrace blocks re-scheduling and sets
 244          * a flag in its per-cpu flags to indicate that it doesn't
 245          * want to fault. On returning from the probe, the no-fault
 246          * flag is cleared and finally re-scheduling is enabled.
 247          */
 248         if ((type == T_PROTFLT || type == T_PAGEFLT) &&
 249             dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
 250                 goto out;
 251 #endif
 252
 253         if ((frame->tf_eflags & PSL_I) == 0) {
 254                 /*
 255                  * Buggy application or kernel code has disabled
 256                  * interrupts and then trapped.  Enabling interrupts
 257                  * now is wrong, but it is better than running with
 258                  * interrupts disabled until they are accidentally
 259                  * enabled later.
 260                  */
 261                 if (ISPL(frame->tf_cs) == SEL_UPL || (frame->tf_eflags & PSL_VM))
 262                         uprintf(
 263                             "pid %ld (%s): trap %d with interrupts disabled\n",
 264                             (long)curproc->p_pid, curthread->td_name, type);
 265                 else if (type != T_NMI && type != T_BPTFLT &&
 266                     type != T_TRCTRAP &&
 267                     frame->tf_eip != (int)cpu_switch_load_gs) {
 268                         /*
 269                          * XXX not quite right, since this may be for a
 270                          * multiple fault in user mode.
 271                          */
 272                         printf("kernel trap %d with interrupts disabled\n",
 273                             type);
 274                         /*
 275                          * Page faults need interrupts disabled until later,
 276                          * and we shouldn't enable interrupts while holding
 277                          * a spin lock.
 278                          */
 279                         if (type != T_PAGEFLT &&
 280                             td->td_md.md_spinlock_count == 0)
 281                                 enable_intr();
 282                 }
 283         }
 284         eva = 0;
 285         code = frame->tf_err;
 286         if (type == T_PAGEFLT) {
 287                 /*
 288                  * For some Cyrix CPUs, %cr2 is clobbered by
 289                  * interrupts.  This problem is worked around by using
 290                  * an interrupt gate for the pagefault handler.  We
 291                  * are finally ready to read %cr2 and conditionally
 292                  * reenable interrupts.  If we hold a spin lock, then
 293                  * we must not reenable interrupts.  This might be a
 294                  * spurious page fault.
 295                  */
 296                 eva = rcr2();
 297                 if (td->td_md.md_spinlock_count == 0)
 298                         enable_intr();
 299         }
 300
 301         if ((ISPL(frame->tf_cs) == SEL_UPL) ||
 302             ((frame->tf_eflags & PSL_VM) &&
 303                 !(curpcb->pcb_flags & PCB_VM86CALL))) {
 304                 /* user trap */
 305
 306                 td->td_pticks = 0;
 307                 td->td_frame = frame;
 308                 addr = frame->tf_eip;
 309                 if (td->td_ucred != p->p_ucred)
 310                         cred_update_thread(td);
 311
 312                 switch (type) {
 313                 case T_PRIVINFLT:       /* privileged instruction fault */
 314                         i = SIGILL;
 315                         ucode = ILL_PRVOPC;
 316                         break;
 317
 318                 case T_BPTFLT:          /* bpt instruction fault */
 319                 case T_TRCTRAP:         /* trace trap */
 320                         enable_intr();
 321 #ifdef KDTRACE_HOOKS
 322                         if (type == T_BPTFLT) {
 323                                 fill_frame_regs(frame, &regs);
 324                                 if (dtrace_pid_probe_ptr != NULL &&
 325                                     dtrace_pid_probe_ptr(&regs) == 0)
 326                                         goto out;
 327                         }
 328 #endif
 329                         frame->tf_eflags &= ~PSL_T;
 330                         i = SIGTRAP;
 331                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 332                         break;
 333
 334                 case T_ARITHTRAP:       /* arithmetic trap */
 335 #ifdef DEV_NPX
 336                         ucode = npxtrap_x87();
 337                         if (ucode == -1)
 338                                 goto userout;
 339 #else
 340                         ucode = 0;
 341 #endif
 342                         i = SIGFPE;
 343                         break;
 344
 345                         /*
 346                          * The following two traps can happen in
 347                          * vm86 mode, and, if so, we want to handle
 348                          * them specially.
 349                          */
 350                 case T_PROTFLT:         /* general protection fault */
 351                 case T_STKFLT:          /* stack fault */
 352                         if (frame->tf_eflags & PSL_VM) {
 353                                 i = vm86_emulate((struct vm86frame *)frame);
 354                                 if (i == 0)
 355                                         goto user;
 356                                 break;
 357                         }
 358                         i = SIGBUS;
 359                         ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
 360                         break;
 361                 case T_SEGNPFLT:        /* segment not present fault */
 362                         i = SIGBUS;
 363                         ucode = BUS_ADRERR;
 364                         break;
 365                 case T_TSSFLT:          /* invalid TSS fault */
 366                         i = SIGBUS;
 367                         ucode = BUS_OBJERR;
 368                         break;
 369                 case T_DOUBLEFLT:       /* double fault */
 370                 default:
 371                         i = SIGBUS;
 372                         ucode = BUS_OBJERR;
 373                         break;
 374
 375                 case T_PAGEFLT:         /* page fault */
 376
 377                         i = trap_pfault(frame, TRUE, eva);
 378 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 379                         if (i == -2) {
 380                                 /*
 381                                  * The f00f hack workaround has triggered, so
 382                                  * treat the fault as an illegal instruction
 383                                  * (T_PRIVINFLT) instead of a page fault.
 384                                  */
 385                                 type = frame->tf_trapno = T_PRIVINFLT;
 386
 387                                 /* Proceed as in that case. */
 388                                 ucode = ILL_PRVOPC;
 389                                 i = SIGILL;
 390                                 break;
 391                         }
 392 #endif
 393                         if (i == -1)
 394                                 goto userout;
 395                         if (i == 0)
 396                                 goto user;
 397
 398                         if (i == SIGSEGV)
 399                                 ucode = SEGV_MAPERR;
 400                         else {
 401                                 if (prot_fault_translation == 0) {
 402                                         /*
 403                                          * Autodetect.
 404                                          * This check also covers the images
 405                                          * without the ABI-tag ELF note.
 406                                          */
 407                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 408                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 409                                                 i = SIGSEGV;
 410                                                 ucode = SEGV_ACCERR;
 411                                         } else {
 412                                                 i = SIGBUS;
 413                                                 ucode = BUS_PAGE_FAULT;
 414                                         }
 415                                 } else if (prot_fault_translation == 1) {
 416                                         /*
 417                                          * Always compat mode.
 418                                          */
 419                                         i = SIGBUS;
 420                                         ucode = BUS_PAGE_FAULT;
 421                                 } else {
 422                                         /*
 423                                          * Always SIGSEGV mode.
 424                                          */
 425                                         i = SIGSEGV;
 426                                         ucode = SEGV_ACCERR;
 427                                 }
 428                         }
 429                         addr = eva;
 430                         break;
 431
 432                 case T_DIVIDE:          /* integer divide fault */
 433                         ucode = FPE_INTDIV;
 434                         i = SIGFPE;
 435                         break;
 436
 437 #ifdef DEV_ISA
 438                 case T_NMI:
 439 #ifdef POWERFAIL_NMI
 440 #ifndef TIMER_FREQ
 441 #  define TIMER_FREQ 1193182
 442 #endif
 443                         if (time_second - lastalert > 10) {
 444                                 log(LOG_WARNING, "NMI: power fail\n");
 445                                 sysbeep(880, hz);
 446                                 lastalert = time_second;
 447                         }
 448                         goto userout;
 449 #else /* !POWERFAIL_NMI */
 450                         /* machine/parity/power fail/"kitchen sink" faults */
 451                         if (isa_nmi(code) == 0) {
 452 #ifdef KDB
 453                                 /*
 454                                  * NMI can be hooked up to a pushbutton
 455                                  * for debugging.
 456                                  */
 457                                 if (kdb_on_nmi) {
 458                                         printf ("NMI ... going to debugger\n");
 459                                         kdb_trap(type, 0, frame);
 460                                 }
 461 #endif /* KDB */
 462                                 goto userout;
 463                         } else if (panic_on_nmi)
 464                                 panic("NMI indicates hardware failure");
 465                         break;
 466 #endif /* POWERFAIL_NMI */
 467 #endif /* DEV_ISA */
 468
 469                 case T_OFLOW:           /* integer overflow fault */
 470                         ucode = FPE_INTOVF;
 471                         i = SIGFPE;
 472                         break;
 473
 474                 case T_BOUND:           /* bounds check fault */
 475                         ucode = FPE_FLTSUB;
 476                         i = SIGFPE;
 477                         break;
 478
 479                 case T_DNA:
 480 #ifdef DEV_NPX
 481                         KASSERT(PCB_USER_FPU(td->td_pcb),
 482                             ("kernel FPU ctx has leaked"));
 483                         /* transparent fault (due to context switch "late") */
 484                         if (npxdna())
 485                                 goto userout;
 486 #endif
 487                         uprintf("pid %d killed due to lack of floating point\n",
 488                                 p->p_pid);
 489                         i = SIGKILL;
 490                         ucode = 0;
 491                         break;
 492
 493                 case T_FPOPFLT:         /* FPU operand fetch fault */
 494                         ucode = ILL_COPROC;
 495                         i = SIGILL;
 496                         break;
 497
 498                 case T_XMMFLT:          /* SIMD floating-point exception */
 499 #if defined(DEV_NPX) && !defined(CPU_DISABLE_SSE) && defined(I686_CPU)
 500                         ucode = npxtrap_sse();
 501                         if (ucode == -1)
 502                                 goto userout;
 503 #else
 504                         ucode = 0;
 505 #endif
 506                         i = SIGFPE;
 507                         break;
 508 #ifdef KDTRACE_HOOKS
 509                 case T_DTRACE_RET:
 510                         enable_intr();
 511                         fill_frame_regs(frame, &regs);
 512                         if (dtrace_return_probe_ptr != NULL &&
 513                             dtrace_return_probe_ptr(&regs) == 0)
 514                                 goto out;
 515                         break;
 516 #endif
 517                 }
 518         } else {
 519                 /* kernel trap */
 520
 521                 KASSERT(cold || td->td_ucred != NULL,
 522                     ("kernel trap doesn't have ucred"));
 523                 switch (type) {
 524                 case T_PAGEFLT:                 /* page fault */
 525                         (void) trap_pfault(frame, FALSE, eva);
 526                         goto out;
 527
 528                 case T_DNA:
 529 #ifdef DEV_NPX
 530                         KASSERT(!PCB_USER_FPU(td->td_pcb),
 531                             ("Unregistered use of FPU in kernel"));
 532                         if (npxdna())
 533                                 goto out;
 534 #endif
 535                         break;
 536
 537                 case T_ARITHTRAP:       /* arithmetic trap */
 538                 case T_XMMFLT:          /* SIMD floating-point exception */
 539                 case T_FPOPFLT:         /* FPU operand fetch fault */
 540                         /*
 541                          * XXXKIB for now disable any FPU traps in kernel
 542                          * handler registration seems to be overkill
 543                          */
 544                         trap_fatal(frame, 0);
 545                         goto out;
 546
 547                         /*
 548                          * The following two traps can happen in
 549                          * vm86 mode, and, if so, we want to handle
 550                          * them specially.
 551                          */
 552                 case T_PROTFLT:         /* general protection fault */
 553                 case T_STKFLT:          /* stack fault */
 554                         if (frame->tf_eflags & PSL_VM) {
 555                                 i = vm86_emulate((struct vm86frame *)frame);
 556                                 if (i != 0)
 557                                         /*
 558                                          * returns to original process
 559                                          */
 560                                         vm86_trap((struct vm86frame *)frame);
 561                                 goto out;
 562                         }
 563                         if (type == T_STKFLT)
 564                                 break;
 565
 566                         /* FALL THROUGH */
 567
 568                 case T_SEGNPFLT:        /* segment not present fault */
 569                         if (curpcb->pcb_flags & PCB_VM86CALL)
 570                                 break;
 571
 572                         /*
 573                          * Invalid %fs's and %gs's can be created using
 574                          * procfs or PT_SETREGS or by invalidating the
 575                          * underlying LDT entry.  This causes a fault
 576                          * in kernel mode when the kernel attempts to
 577                          * switch contexts.  Lose the bad context
 578                          * (XXX) so that we can continue, and generate
 579                          * a signal.
 580                          */
 581                         if (frame->tf_eip == (int)cpu_switch_load_gs) {
 582                                 curpcb->pcb_gs = 0;
 583 #if 0
 584                                 PROC_LOCK(p);
 585                                 kern_psignal(p, SIGBUS);
 586                                 PROC_UNLOCK(p);
 587 #endif
 588                                 goto out;
 589                         }
 590
 591                         if (td->td_intr_nesting_level != 0)
 592                                 break;
 593
 594                         /*
 595                          * Invalid segment selectors and out of bounds
 596                          * %eip's and %esp's can be set up in user mode.
 597                          * This causes a fault in kernel mode when the
 598                          * kernel tries to return to user mode.  We want
 599                          * to get this fault so that we can fix the
 600                          * problem here and not have to check all the
 601                          * selectors and pointers when the user changes
 602                          * them.
 603                          */
 604                         if (frame->tf_eip == (int)doreti_iret) {
 605                                 frame->tf_eip = (int)doreti_iret_fault;
 606                                 goto out;
 607                         }
 608                         if (frame->tf_eip == (int)doreti_popl_ds) {
 609                                 frame->tf_eip = (int)doreti_popl_ds_fault;
 610                                 goto out;
 611                         }
 612                         if (frame->tf_eip == (int)doreti_popl_es) {
 613                                 frame->tf_eip = (int)doreti_popl_es_fault;
 614                                 goto out;
 615                         }
 616                         if (frame->tf_eip == (int)doreti_popl_fs) {
 617                                 frame->tf_eip = (int)doreti_popl_fs_fault;
 618                                 goto out;
 619                         }
 620                         if (curpcb->pcb_onfault != NULL) {
 621                                 frame->tf_eip =
 622                                     (int)curpcb->pcb_onfault;
 623                                 goto out;
 624                         }
 625                         break;
 626
 627                 case T_TSSFLT:
 628                         /*
 629                          * PSL_NT can be set in user mode and isn't cleared
 630                          * automatically when the kernel is entered.  This
 631                          * causes a TSS fault when the kernel attempts to
 632                          * `iret' because the TSS link is uninitialized.  We
 633                          * want to get this fault so that we can fix the
 634                          * problem here and not every time the kernel is
 635                          * entered.
 636                          */
 637                         if (frame->tf_eflags & PSL_NT) {
 638                                 frame->tf_eflags &= ~PSL_NT;
 639                                 goto out;
 640                         }
 641                         break;
 642
 643                 case T_TRCTRAP:  /* trace trap */
 644                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall)) {
 645                                 /*
 646                                  * We've just entered system mode via the
 647                                  * syscall lcall.  Continue single stepping
 648                                  * silently until the syscall handler has
 649                                  * saved the flags.
 650                                  */
 651                                 goto out;
 652                         }
 653                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall) + 1) {
 654                                 /*
 655                                  * The syscall handler has now saved the
 656                                  * flags.  Stop single stepping it.
 657                                  */
 658                                 frame->tf_eflags &= ~PSL_T;
 659                                 goto out;
 660                         }
 661                         /*
 662                          * Ignore debug register trace traps due to
 663                          * accesses in the user's address space, which
 664                          * can happen under several conditions such as
 665                          * if a user sets a watchpoint on a buffer and
 666                          * then passes that buffer to a system call.
 667                          * We still want to get TRCTRAPS for addresses
 668                          * in kernel space because that is useful when
 669                          * debugging the kernel.
 670                          */
 671                         if (user_dbreg_trap() &&
 672                            !(curpcb->pcb_flags & PCB_VM86CALL)) {
 673                                 /*
 674                                  * Reset breakpoint bits because the
 675                                  * processor doesn't
 676                                  */
 677                                 load_dr6(rdr6() & 0xfffffff0);
 678                                 goto out;
 679                         }
 680                         /*
 681                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 682                          */
 683                 case T_BPTFLT:
 684                         /*
 685                          * If KDB is enabled, let it handle the debugger trap.
 686                          * Otherwise, debugger traps "can't happen".
 687                          */
 688 #ifdef KDB
 689                         if (kdb_trap(type, 0, frame))
 690                                 goto out;
 691 #endif
 692                         break;
 693
 694 #ifdef DEV_ISA
 695                 case T_NMI:
 696 #ifdef POWERFAIL_NMI
 697                         if (time_second - lastalert > 10) {
 698                                 log(LOG_WARNING, "NMI: power fail\n");
 699                                 sysbeep(880, hz);
 700                                 lastalert = time_second;
 701                         }
 702                         goto out;
 703 #else /* !POWERFAIL_NMI */
 704                         /* machine/parity/power fail/"kitchen sink" faults */
 705                         if (isa_nmi(code) == 0) {
 706 #ifdef KDB
 707                                 /*
 708                                  * NMI can be hooked up to a pushbutton
 709                                  * for debugging.
 710                                  */
 711                                 if (kdb_on_nmi) {
 712                                         printf ("NMI ... going to debugger\n");
 713                                         kdb_trap(type, 0, frame);
 714                                 }
 715 #endif /* KDB */
 716                                 goto out;
 717                         } else if (panic_on_nmi == 0)
 718                                 goto out;
 719                         /* FALLTHROUGH */
 720 #endif /* POWERFAIL_NMI */
 721 #endif /* DEV_ISA */
 722                 }
 723
 724                 trap_fatal(frame, eva);
 725                 goto out;
 726         }
 727
 728         /* Translate fault for emulators (e.g. Linux) */
 729         if (*p->p_sysent->sv_transtrap)
 730                 i = (*p->p_sysent->sv_transtrap)(i, type);
 731
 732         ksiginfo_init_trap(&ksi);
 733         ksi.ksi_signo = i;
 734         ksi.ksi_code = ucode;
 735         ksi.ksi_addr = (void *)addr;
 736         ksi.ksi_trapno = type;
 737         if (uprintf_signal) {
 738                 uprintf("pid %d comm %s: signal %d err %x code %d type %d "
 739                     "addr 0x%x esp 0x%08x eip 0x%08x "
 740                     "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
 741                     p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr,
 742                     frame->tf_esp, frame->tf_eip,
 743                     fubyte((void *)(frame->tf_eip + 0)),
 744                     fubyte((void *)(frame->tf_eip + 1)),
 745                     fubyte((void *)(frame->tf_eip + 2)),
 746                     fubyte((void *)(frame->tf_eip + 3)),
 747                     fubyte((void *)(frame->tf_eip + 4)),
 748                     fubyte((void *)(frame->tf_eip + 5)),
 749                     fubyte((void *)(frame->tf_eip + 6)),
 750                     fubyte((void *)(frame->tf_eip + 7)));
 751         }
 752         KASSERT((read_eflags() & PSL_I) != 0, ("interrupts disabled"));
 753         trapsignal(td, &ksi);
 754
 755 #ifdef DEBUG
 756         if (type <= MAX_TRAP_MSG) {
 757                 uprintf("fatal process exception: %s",
 758                         trap_msg[type]);
 759                 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
 760                         uprintf(", fault VA = 0x%lx", (u_long)eva);
 761                 uprintf("\n");
 762         }
 763 #endif
 764
 765 user:
 766         userret(td, frame);
 767         KASSERT(PCB_USER_FPU(td->td_pcb),
 768             ("Return from trap with kernel FPU ctx leaked"));
 769 userout:
 770 out:
 771         return;
 772 }
 773
 774 static int
 775 trap_pfault(frame, usermode, eva)
 776         struct trapframe *frame;
 777         int usermode;
 778         vm_offset_t eva;
 779 {
 780         vm_offset_t va;
 781         struct vmspace *vm;
 782         vm_map_t map;
 783         int rv = 0;
 784         vm_prot_t ftype;
 785         struct thread *td = curthread;
 786         struct proc *p = td->td_proc;
 787
 788         if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
 789                 /*
 790                  * Due to both processor errata and lazy TLB invalidation when
 791                  * access restrictions are removed from virtual pages, memory
 792                  * accesses that are allowed by the physical mapping layer may
 793                  * nonetheless cause one spurious page fault per virtual page.
 794                  * When the thread is executing a "no faulting" section that
 795                  * is bracketed by vm_fault_{disable,enable}_pagefaults(),
 796                  * every page fault is treated as a spurious page fault,
 797                  * unless it accesses the same virtual address as the most
 798                  * recent page fault within the same "no faulting" section.
 799                  */
 800                 if (td->td_md.md_spurflt_addr != eva ||
 801                     (td->td_pflags & TDP_RESETSPUR) != 0) {
 802                         /*
 803                          * Do nothing to the TLB.  A stale TLB entry is
 804                          * flushed automatically by a page fault.
 805                          */
 806                         td->td_md.md_spurflt_addr = eva;
 807                         td->td_pflags &= ~TDP_RESETSPUR;
 808                         return (0);
 809                 }
 810         } else {
 811                 /*
 812                  * If we get a page fault while in a critical section, then
 813                  * it is most likely a fatal kernel page fault.  The kernel
 814                  * is already going to panic trying to get a sleep lock to
 815                  * do the VM lookup, so just consider it a fatal trap so the
 816                  * kernel can print out a useful trap message and even get
 817                  * to the debugger.
 818                  *
 819                  * If we get a page fault while holding a non-sleepable
 820                  * lock, then it is most likely a fatal kernel page fault.
 821                  * If WITNESS is enabled, then it's going to whine about
 822                  * bogus LORs with various VM locks, so just skip to the
 823                  * fatal trap handling directly.
 824                  */
 825                 if (td->td_critnest != 0 ||
 826                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 827                     "Kernel page fault") != 0) {
 828                         trap_fatal(frame, eva);
 829                         return (-1);
 830                 }
 831         }
 832         va = trunc_page(eva);
 833         if (va >= KERNBASE) {
 834                 /*
 835                  * Don't allow user-mode faults in kernel address space.
 836                  * An exception:  if the faulting address is the invalid
 837                  * instruction entry in the IDT, then the Intel Pentium
 838                  * F00F bug workaround was triggered, and we need to
 839                  * treat it is as an illegal instruction, and not a page
 840                  * fault.
 841                  */
 842 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 843                 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
 844                         return (-2);
 845 #endif
 846                 if (usermode)
 847                         goto nogo;
 848
 849                 map = kernel_map;
 850         } else {
 851                 /*
 852                  * This is a fault on non-kernel virtual memory.  If either
 853                  * p or p->p_vmspace is NULL, then the fault is fatal.
 854                  */
 855                 if (p == NULL || (vm = p->p_vmspace) == NULL)
 856                         goto nogo;
 857
 858                 map = &vm->vm_map;
 859
 860                 /*
 861                  * When accessing a user-space address, kernel must be
 862                  * ready to accept the page fault, and provide a
 863                  * handling routine.  Since accessing the address
 864                  * without the handler is a bug, do not try to handle
 865                  * it normally, and panic immediately.
 866                  */
 867                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 868                     curpcb->pcb_onfault == NULL)) {
 869                         trap_fatal(frame, eva);
 870                         return (-1);
 871                 }
 872         }
 873
 874         /*
 875          * PGEX_I is defined only if the execute disable bit capability is
 876          * supported and enabled.
 877          */
 878         if (frame->tf_err & PGEX_W)
 879                 ftype = VM_PROT_WRITE;
 880 #ifdef PAE
 881         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 882                 ftype = VM_PROT_EXECUTE;
 883 #endif
 884         else
 885                 ftype = VM_PROT_READ;
 886
 887         if (map != kernel_map) {
 888                 /*
 889                  * Keep swapout from messing with us during this
 890                  *      critical time.
 891                  */
 892                 PROC_LOCK(p);
 893                 ++p->p_lock;
 894                 PROC_UNLOCK(p);
 895
 896                 /* Fault in the user page: */
 897                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 898
 899                 PROC_LOCK(p);
 900                 --p->p_lock;
 901                 PROC_UNLOCK(p);
 902         } else {
 903                 /*
 904                  * Don't have to worry about process locking or stacks in the
 905                  * kernel.
 906                  */
 907                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 908         }
 909         if (rv == KERN_SUCCESS) {
 910 #ifdef HWPMC_HOOKS
 911                 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
 912                         PMC_SOFT_CALL_TF( , , page_fault, all, frame);
 913                         if (ftype == VM_PROT_READ)
 914                                 PMC_SOFT_CALL_TF( , , page_fault, read,
 915                                     frame);
 916                         else
 917                                 PMC_SOFT_CALL_TF( , , page_fault, write,
 918                                     frame);
 919                 }
 920 #endif
 921                 return (0);
 922         }
 923 nogo:
 924         if (!usermode) {
 925                 if (td->td_intr_nesting_level == 0 &&
 926                     curpcb->pcb_onfault != NULL) {
 927                         frame->tf_eip = (int)curpcb->pcb_onfault;
 928                         return (0);
 929                 }
 930                 trap_fatal(frame, eva);
 931                 return (-1);
 932         }
 933         return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 934 }
 935
 936 static void
 937 trap_fatal(frame, eva)
 938         struct trapframe *frame;
 939         vm_offset_t eva;
 940 {
 941         int code, ss, esp;
 942         u_int type;
 943         struct soft_segment_descriptor softseg;
 944         char *msg;
 945
 946         code = frame->tf_err;
 947         type = frame->tf_trapno;
 948         sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
 949
 950         if (type <= MAX_TRAP_MSG)
 951                 msg = trap_msg[type];
 952         else
 953                 msg = "UNKNOWN";
 954         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 955             frame->tf_eflags & PSL_VM ? "vm86" :
 956             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
 957 #ifdef SMP
 958         /* two separate prints in case of a trap on an unmapped page */
 959         printf("cpuid = %d; ", PCPU_GET(cpuid));
 960         printf("apic id = %02x\n", PCPU_GET(apic_id));
 961 #endif
 962         if (type == T_PAGEFLT) {
 963                 printf("fault virtual address   = 0x%x\n", eva);
 964                 printf("fault code              = %s %s, %s\n",
 965                         code & PGEX_U ? "user" : "supervisor",
 966                         code & PGEX_W ? "write" : "read",
 967                         code & PGEX_P ? "protection violation" : "page not present");
 968         }
 969         printf("instruction pointer     = 0x%x:0x%x\n",
 970                frame->tf_cs & 0xffff, frame->tf_eip);
 971         if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
 972                 ss = frame->tf_ss & 0xffff;
 973                 esp = frame->tf_esp;
 974         } else {
 975                 ss = GSEL(GDATA_SEL, SEL_KPL);
 976                 esp = (int)&frame->tf_esp;
 977         }
 978         printf("stack pointer           = 0x%x:0x%x\n", ss, esp);
 979         printf("frame pointer           = 0x%x:0x%x\n", ss, frame->tf_ebp);
 980         printf("code segment            = base 0x%x, limit 0x%x, type 0x%x\n",
 981                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 982         printf("                        = DPL %d, pres %d, def32 %d, gran %d\n",
 983                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
 984                softseg.ssd_gran);
 985         printf("processor eflags        = ");
 986         if (frame->tf_eflags & PSL_T)
 987                 printf("trace trap, ");
 988         if (frame->tf_eflags & PSL_I)
 989                 printf("interrupt enabled, ");
 990         if (frame->tf_eflags & PSL_NT)
 991                 printf("nested task, ");
 992         if (frame->tf_eflags & PSL_RF)
 993                 printf("resume, ");
 994         if (frame->tf_eflags & PSL_VM)
 995                 printf("vm86, ");
 996         printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
 997         printf("current process         = ");
 998         if (curproc) {
 999                 printf("%lu (%s)\n", (u_long)curproc->p_pid, curthread->td_name);
1000         } else {
1001                 printf("Idle\n");
1002         }
1003
1004 #ifdef KDB
1005         if (debugger_on_panic || kdb_active) {
1006                 frame->tf_err = eva;    /* smuggle fault address to ddb */
1007                 if (kdb_trap(type, 0, frame)) {
1008                         frame->tf_err = code;   /* restore error code */
1009                         return;
1010                 }
1011                 frame->tf_err = code;           /* restore error code */
1012         }
1013 #endif
1014         printf("trap number             = %d\n", type);
1015         if (type <= MAX_TRAP_MSG)
1016                 panic("%s", trap_msg[type]);
1017         else
1018                 panic("unknown/reserved trap");
1019 }
1020
1021 /*
1022  * Double fault handler. Called when a fault occurs while writing
1023  * a frame for a trap/exception onto the stack. This usually occurs
1024  * when the stack overflows (such is the case with infinite recursion,
1025  * for example).
1026  *
1027  * XXX Note that the current PTD gets replaced by IdlePTD when the
1028  * task switch occurs. This means that the stack that was active at
1029  * the time of the double fault is not available at <kstack> unless
1030  * the machine was idle when the double fault occurred. The downside
1031  * of this is that "trace <ebp>" in ddb won't work.
1032  */
1033 void
1034 dblfault_handler()
1035 {
1036 #ifdef KDTRACE_HOOKS
1037         if (dtrace_doubletrap_func != NULL)
1038                 (*dtrace_doubletrap_func)();
1039 #endif
1040         printf("\nFatal double fault:\n");
1041         printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip));
1042         printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp));
1043         printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp));
1044 #ifdef SMP
1045         /* two separate prints in case of a trap on an unmapped page */
1046         printf("cpuid = %d; ", PCPU_GET(cpuid));
1047         printf("apic id = %02x\n", PCPU_GET(apic_id));
1048 #endif
1049         panic("double fault");
1050 }
1051
1052 int
1053 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
1054 {
1055         struct proc *p;
1056         struct trapframe *frame;
1057         caddr_t params;
1058         int error;
1059
1060         p = td->td_proc;
1061         frame = td->td_frame;
1062
1063         params = (caddr_t)frame->tf_esp + sizeof(int);
1064         sa->code = frame->tf_eax;
1065
1066         /*
1067          * Need to check if this is a 32 bit or 64 bit syscall.
1068          */
1069         if (sa->code == SYS_syscall) {
1070                 /*
1071                  * Code is first argument, followed by actual args.
1072                  */
1073                 sa->code = fuword(params);
1074                 params += sizeof(int);
1075         } else if (sa->code == SYS___syscall) {
1076                 /*
1077                  * Like syscall, but code is a quad, so as to maintain
1078                  * quad alignment for the rest of the arguments.
1079                  */
1080                 sa->code = fuword(params);
1081                 params += sizeof(quad_t);
1082         }
1083
1084         if (p->p_sysent->sv_mask)
1085                 sa->code &= p->p_sysent->sv_mask;
1086         if (sa->code >= p->p_sysent->sv_size)
1087                 sa->callp = &p->p_sysent->sv_table[0];
1088         else
1089                 sa->callp = &p->p_sysent->sv_table[sa->code];
1090         sa->narg = sa->callp->sy_narg;
1091
1092         if (params != NULL && sa->narg != 0)
1093                 error = copyin(params, (caddr_t)sa->args,
1094                     (u_int)(sa->narg * sizeof(int)));
1095         else
1096                 error = 0;
1097
1098         if (error == 0) {
1099                 td->td_retval[0] = 0;
1100                 td->td_retval[1] = frame->tf_edx;
1101         }
1102
1103         return (error);
1104 }
1105
1106 #include "../../kern/subr_syscall.c"
1107
1108 /*
1109  * syscall - system call request C handler.  A system call is
1110  * essentially treated as a trap by reusing the frame layout.
1111  */
1112 void
1113 syscall(struct trapframe *frame)
1114 {
1115         struct thread *td;
1116         struct syscall_args sa;
1117         register_t orig_tf_eflags;
1118         int error;
1119         ksiginfo_t ksi;
1120
1121 #ifdef DIAGNOSTIC
1122         if (ISPL(frame->tf_cs) != SEL_UPL) {
1123                 panic("syscall");
1124                 /* NOT REACHED */
1125         }
1126 #endif
1127         orig_tf_eflags = frame->tf_eflags;
1128
1129         td = curthread;
1130         td->td_frame = frame;
1131
1132         error = syscallenter(td, &sa);
1133
1134         /*
1135          * Traced syscall.
1136          */
1137         if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1138                 frame->tf_eflags &= ~PSL_T;
1139                 ksiginfo_init_trap(&ksi);
1140                 ksi.ksi_signo = SIGTRAP;
1141                 ksi.ksi_code = TRAP_TRACE;
1142                 ksi.ksi_addr = (void *)frame->tf_eip;
1143                 trapsignal(td, &ksi);
1144         }
1145
1146         KASSERT(PCB_USER_FPU(td->td_pcb),
1147             ("System call %s returning with kernel FPU ctx leaked",
1148              syscallname(td->td_proc, sa.code)));
1149         KASSERT(td->td_pcb->pcb_save == &td->td_pcb->pcb_user_save,
1150             ("System call %s returning with mangled pcb_save",
1151              syscallname(td->td_proc, sa.code)));
1152
1153         syscallret(td, error, &sa);
1154 }