sys/i386/i386/trap.c

   1 /*-
   2  * Copyright (C) 1994, David Greenman
   3  * Copyright (c) 1990, 1993
   4  *      The Regents of the University of California.  All rights reserved.
   5  *
   6  * This code is derived from software contributed to Berkeley by
   7  * the University of Utah, and William Jolitz.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in the
  16  *    documentation and/or other materials provided with the distribution.
  17  * 3. All advertising materials mentioning features or use of this software
  18  *    must display the following acknowledgement:
  19  *      This product includes software developed by the University of
  20  *      California, Berkeley and its contributors.
  21  * 4. Neither the name of the University nor the names of its contributors
  22  *    may be used to endorse or promote products derived from this software
  23  *    without specific prior written permission.
  24  *
  25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  35  * SUCH DAMAGE.
  36  *
  37  *      from: @(#)trap.c        7.4 (Berkeley) 5/13/91
  38  */
  39
  40 #include <sys/cdefs.h>
  41 __FBSDID("$FreeBSD$");
  42
  43 /*
  44  * 386 Trap and System call handling
  45  */
  46
  47 #include "opt_clock.h"
  48 #include "opt_cpu.h"
  49 #include "opt_hwpmc_hooks.h"
  50 #include "opt_isa.h"
  51 #include "opt_kdb.h"
  52 #include "opt_kdtrace.h"
  53 #include "opt_npx.h"
  54 #include "opt_trap.h"
  55
  56 #include <sys/param.h>
  57 #include <sys/bus.h>
  58 #include <sys/systm.h>
  59 #include <sys/proc.h>
  60 #include <sys/pioctl.h>
  61 #include <sys/ptrace.h>
  62 #include <sys/kdb.h>
  63 #include <sys/kernel.h>
  64 #include <sys/ktr.h>
  65 #include <sys/lock.h>
  66 #include <sys/mutex.h>
  67 #include <sys/resourcevar.h>
  68 #include <sys/signalvar.h>
  69 #include <sys/syscall.h>
  70 #include <sys/sysctl.h>
  71 #include <sys/sysent.h>
  72 #include <sys/uio.h>
  73 #include <sys/vmmeter.h>
  74 #ifdef HWPMC_HOOKS
  75 #include <sys/pmckern.h>
  76 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/tss.h>
  99 #include <machine/vm86.h>
 100
 101 #ifdef POWERFAIL_NMI
 102 #include <sys/syslog.h>
 103 #include <machine/clock.h>
 104 #endif
 105
 106 #ifdef KDTRACE_HOOKS
 107 #include <sys/dtrace_bsd.h>
 108
 109 /*
 110  * This is a hook which is initialised by the dtrace module
 111  * to handle traps which might occur during DTrace probe
 112  * execution.
 113  */
 114 dtrace_trap_func_t      dtrace_trap_func;
 115
 116 dtrace_doubletrap_func_t        dtrace_doubletrap_func;
 117
 118 /*
 119  * This is a hook which is initialised by the systrace module
 120  * when it is loaded. This keeps the DTrace syscall provider
 121  * implementation opaque.
 122  */
 123 systrace_probe_func_t   systrace_probe_func;
 124
 125 /*
 126  * These hooks are necessary for the pid and usdt providers.
 127  */
 128 dtrace_pid_probe_ptr_t          dtrace_pid_probe_ptr;
 129 dtrace_return_probe_ptr_t       dtrace_return_probe_ptr;
 130 #endif
 131
 132 extern void trap(struct trapframe *frame);
 133 extern void syscall(struct trapframe *frame);
 134
 135 static int trap_pfault(struct trapframe *, int, vm_offset_t);
 136 static void trap_fatal(struct trapframe *, vm_offset_t);
 137 void dblfault_handler(void);
 138
 139 extern inthand_t IDTVEC(lcall_syscall);
 140
 141 #define MAX_TRAP_MSG            32
 142 static char *trap_msg[] = {
 143         "",                                     /*  0 unused */
 144         "privileged instruction fault",         /*  1 T_PRIVINFLT */
 145         "",                                     /*  2 unused */
 146         "breakpoint instruction fault",         /*  3 T_BPTFLT */
 147         "",                                     /*  4 unused */
 148         "",                                     /*  5 unused */
 149         "arithmetic trap",                      /*  6 T_ARITHTRAP */
 150         "",                                     /*  7 unused */
 151         "",                                     /*  8 unused */
 152         "general protection fault",             /*  9 T_PROTFLT */
 153         "trace trap",                           /* 10 T_TRCTRAP */
 154         "",                                     /* 11 unused */
 155         "page fault",                           /* 12 T_PAGEFLT */
 156         "",                                     /* 13 unused */
 157         "alignment fault",                      /* 14 T_ALIGNFLT */
 158         "",                                     /* 15 unused */
 159         "",                                     /* 16 unused */
 160         "",                                     /* 17 unused */
 161         "integer divide fault",                 /* 18 T_DIVIDE */
 162         "non-maskable interrupt trap",          /* 19 T_NMI */
 163         "overflow trap",                        /* 20 T_OFLOW */
 164         "FPU bounds check fault",               /* 21 T_BOUND */
 165         "FPU device not available",             /* 22 T_DNA */
 166         "double fault",                         /* 23 T_DOUBLEFLT */
 167         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
 168         "invalid TSS fault",                    /* 25 T_TSSFLT */
 169         "segment not present fault",            /* 26 T_SEGNPFLT */
 170         "stack fault",                          /* 27 T_STKFLT */
 171         "machine check trap",                   /* 28 T_MCHK */
 172         "SIMD floating-point exception",        /* 29 T_XMMFLT */
 173         "reserved (unknown) fault",             /* 30 T_RESERVED */
 174         "",                                     /* 31 unused (reserved) */
 175         "DTrace pid return trap",               /* 32 T_DTRACE_RET */
 176 };
 177
 178 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 179 extern int has_f00f_bug;
 180 #endif
 181
 182 #ifdef KDB
 183 static int kdb_on_nmi = 1;
 184 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
 185         &kdb_on_nmi, 0, "Go to KDB on NMI");
 186 TUNABLE_INT("machdep.kdb_on_nmi", &kdb_on_nmi);
 187 #endif
 188 static int panic_on_nmi = 1;
 189 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
 190         &panic_on_nmi, 0, "Panic on NMI");
 191 TUNABLE_INT("machdep.panic_on_nmi", &panic_on_nmi);
 192 static int prot_fault_translation = 0;
 193 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
 194         &prot_fault_translation, 0, "Select signal to deliver on protection fault");
 195 static int uprintf_signal;
 196 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW,
 197     &uprintf_signal, 0,
 198     "Print debugging information on trap signal to ctty");
 199
 200 /*
 201  * Exception, fault, and trap interface to the FreeBSD kernel.
 202  * This common code is called from assembly language IDT gate entry
 203  * routines that prepare a suitable stack frame, and restore this
 204  * frame after the exception has been processed.
 205  */
 206
 207 void
 208 trap(struct trapframe *frame)
 209 {
 210 #ifdef KDTRACE_HOOKS
 211         struct reg regs;
 212 #endif
 213         struct thread *td = curthread;
 214         struct proc *p = td->td_proc;
 215         int i = 0, ucode = 0, code;
 216         u_int type;
 217         register_t addr = 0;
 218         vm_offset_t eva;
 219         ksiginfo_t ksi;
 220 #ifdef POWERFAIL_NMI
 221         static int lastalert = 0;
 222 #endif
 223
 224         PCPU_INC(cnt.v_trap);
 225         type = frame->tf_trapno;
 226
 227 #ifdef SMP
 228         /* Handler for NMI IPIs used for stopping CPUs. */
 229         if (type == T_NMI) {
 230                  if (ipi_nmi_handler() == 0)
 231                            goto out;
 232         }
 233 #endif /* SMP */
 234
 235 #ifdef KDB
 236         if (kdb_active) {
 237                 kdb_reenter();
 238                 goto out;
 239         }
 240 #endif
 241
 242         if (type == T_RESERVED) {
 243                 trap_fatal(frame, 0);
 244                 goto out;
 245         }
 246
 247 #ifdef  HWPMC_HOOKS
 248         /*
 249          * CPU PMCs interrupt using an NMI so we check for that first.
 250          * If the HWPMC module is active, 'pmc_hook' will point to
 251          * the function to be called.  A return value of '1' from the
 252          * hook means that the NMI was handled by it and that we can
 253          * return immediately.
 254          */
 255         if (type == T_NMI && pmc_intr &&
 256             (*pmc_intr)(PCPU_GET(cpuid), frame))
 257             goto out;
 258 #endif
 259
 260         if (type == T_MCHK) {
 261                 mca_intr();
 262                 goto out;
 263         }
 264
 265 #ifdef KDTRACE_HOOKS
 266         /*
 267          * A trap can occur while DTrace executes a probe. Before
 268          * executing the probe, DTrace blocks re-scheduling and sets
 269          * a flag in its per-cpu flags to indicate that it doesn't
 270          * want to fault. On returning from the probe, the no-fault
 271          * flag is cleared and finally re-scheduling is enabled.
 272          */
 273         if ((type == T_PROTFLT || type == T_PAGEFLT) &&
 274             dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
 275                 goto out;
 276 #endif
 277
 278         if ((frame->tf_eflags & PSL_I) == 0) {
 279                 /*
 280                  * Buggy application or kernel code has disabled
 281                  * interrupts and then trapped.  Enabling interrupts
 282                  * now is wrong, but it is better than running with
 283                  * interrupts disabled until they are accidentally
 284                  * enabled later.
 285                  */
 286                 if (ISPL(frame->tf_cs) == SEL_UPL || (frame->tf_eflags & PSL_VM))
 287                         uprintf(
 288                             "pid %ld (%s): trap %d with interrupts disabled\n",
 289                             (long)curproc->p_pid, curthread->td_name, type);
 290                 else if (type != T_BPTFLT && type != T_TRCTRAP &&
 291                          frame->tf_eip != (int)cpu_switch_load_gs) {
 292                         /*
 293                          * XXX not quite right, since this may be for a
 294                          * multiple fault in user mode.
 295                          */
 296                         printf("kernel trap %d with interrupts disabled\n",
 297                             type);
 298                         /*
 299                          * Page faults need interrupts disabled until later,
 300                          * and we shouldn't enable interrupts while holding
 301                          * a spin lock or if servicing an NMI.
 302                          */
 303                         if (type != T_NMI && type != T_PAGEFLT &&
 304                             td->td_md.md_spinlock_count == 0)
 305                                 enable_intr();
 306                 }
 307         }
 308         eva = 0;
 309         code = frame->tf_err;
 310         if (type == T_PAGEFLT) {
 311                 /*
 312                  * For some Cyrix CPUs, %cr2 is clobbered by
 313                  * interrupts.  This problem is worked around by using
 314                  * an interrupt gate for the pagefault handler.  We
 315                  * are finally ready to read %cr2 and conditionally
 316                  * reenable interrupts.  If we hold a spin lock, then
 317                  * we must not reenable interrupts.  This might be a
 318                  * spurious page fault.
 319                  */
 320                 eva = rcr2();
 321                 if (td->td_md.md_spinlock_count == 0)
 322                         enable_intr();
 323         }
 324
 325         if ((ISPL(frame->tf_cs) == SEL_UPL) ||
 326             ((frame->tf_eflags & PSL_VM) &&
 327                 !(curpcb->pcb_flags & PCB_VM86CALL))) {
 328                 /* user trap */
 329
 330                 td->td_pticks = 0;
 331                 td->td_frame = frame;
 332                 addr = frame->tf_eip;
 333                 if (td->td_ucred != p->p_ucred)
 334                         cred_update_thread(td);
 335
 336                 switch (type) {
 337                 case T_PRIVINFLT:       /* privileged instruction fault */
 338                         i = SIGILL;
 339                         ucode = ILL_PRVOPC;
 340                         break;
 341
 342                 case T_BPTFLT:          /* bpt instruction fault */
 343                 case T_TRCTRAP:         /* trace trap */
 344                         enable_intr();
 345 #ifdef KDTRACE_HOOKS
 346                         if (type == T_BPTFLT) {
 347                                 fill_frame_regs(frame, &regs);
 348                                 if (dtrace_pid_probe_ptr != NULL &&
 349                                     dtrace_pid_probe_ptr(&regs) == 0)
 350                                         goto out;
 351                         }
 352 #endif
 353                         frame->tf_eflags &= ~PSL_T;
 354                         i = SIGTRAP;
 355                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 356                         break;
 357
 358                 case T_ARITHTRAP:       /* arithmetic trap */
 359 #ifdef DEV_NPX
 360                         ucode = npxtrap_x87();
 361                         if (ucode == -1)
 362                                 goto userout;
 363 #else
 364                         ucode = 0;
 365 #endif
 366                         i = SIGFPE;
 367                         break;
 368
 369                         /*
 370                          * The following two traps can happen in
 371                          * vm86 mode, and, if so, we want to handle
 372                          * them specially.
 373                          */
 374                 case T_PROTFLT:         /* general protection fault */
 375                 case T_STKFLT:          /* stack fault */
 376                         if (frame->tf_eflags & PSL_VM) {
 377                                 i = vm86_emulate((struct vm86frame *)frame);
 378                                 if (i == 0)
 379                                         goto user;
 380                                 break;
 381                         }
 382                         i = SIGBUS;
 383                         ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
 384                         break;
 385                 case T_SEGNPFLT:        /* segment not present fault */
 386                         i = SIGBUS;
 387                         ucode = BUS_ADRERR;
 388                         break;
 389                 case T_TSSFLT:          /* invalid TSS fault */
 390                         i = SIGBUS;
 391                         ucode = BUS_OBJERR;
 392                         break;
 393                 case T_ALIGNFLT:
 394                         i = SIGBUS;
 395                         ucode = BUS_ADRALN;
 396                         break;
 397                 case T_DOUBLEFLT:       /* double fault */
 398                 default:
 399                         i = SIGBUS;
 400                         ucode = BUS_OBJERR;
 401                         break;
 402
 403                 case T_PAGEFLT:         /* page fault */
 404
 405                         i = trap_pfault(frame, TRUE, eva);
 406 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 407                         if (i == -2) {
 408                                 /*
 409                                  * The f00f hack workaround has triggered, so
 410                                  * treat the fault as an illegal instruction
 411                                  * (T_PRIVINFLT) instead of a page fault.
 412                                  */
 413                                 type = frame->tf_trapno = T_PRIVINFLT;
 414
 415                                 /* Proceed as in that case. */
 416                                 ucode = ILL_PRVOPC;
 417                                 i = SIGILL;
 418                                 break;
 419                         }
 420 #endif
 421                         if (i == -1)
 422                                 goto userout;
 423                         if (i == 0)
 424                                 goto user;
 425
 426                         if (i == SIGSEGV)
 427                                 ucode = SEGV_MAPERR;
 428                         else {
 429                                 if (prot_fault_translation == 0) {
 430                                         /*
 431                                          * Autodetect.
 432                                          * This check also covers the images
 433                                          * without the ABI-tag ELF note.
 434                                          */
 435                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 436                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 437                                                 i = SIGSEGV;
 438                                                 ucode = SEGV_ACCERR;
 439                                         } else {
 440                                                 i = SIGBUS;
 441                                                 ucode = BUS_PAGE_FAULT;
 442                                         }
 443                                 } else if (prot_fault_translation == 1) {
 444                                         /*
 445                                          * Always compat mode.
 446                                          */
 447                                         i = SIGBUS;
 448                                         ucode = BUS_PAGE_FAULT;
 449                                 } else {
 450                                         /*
 451                                          * Always SIGSEGV mode.
 452                                          */
 453                                         i = SIGSEGV;
 454                                         ucode = SEGV_ACCERR;
 455                                 }
 456                         }
 457                         addr = eva;
 458                         break;
 459
 460                 case T_DIVIDE:          /* integer divide fault */
 461                         ucode = FPE_INTDIV;
 462                         i = SIGFPE;
 463                         break;
 464
 465 #ifdef DEV_ISA
 466                 case T_NMI:
 467 #ifdef POWERFAIL_NMI
 468 #ifndef TIMER_FREQ
 469 #  define TIMER_FREQ 1193182
 470 #endif
 471                         if (time_second - lastalert > 10) {
 472                                 log(LOG_WARNING, "NMI: power fail\n");
 473                                 sysbeep(880, hz);
 474                                 lastalert = time_second;
 475                         }
 476                         goto userout;
 477 #else /* !POWERFAIL_NMI */
 478                         /* machine/parity/power fail/"kitchen sink" faults */
 479                         if (isa_nmi(code) == 0) {
 480 #ifdef KDB
 481                                 /*
 482                                  * NMI can be hooked up to a pushbutton
 483                                  * for debugging.
 484                                  */
 485                                 if (kdb_on_nmi) {
 486                                         printf ("NMI ... going to debugger\n");
 487                                         kdb_trap(type, 0, frame);
 488                                 }
 489 #endif /* KDB */
 490                                 goto userout;
 491                         } else if (panic_on_nmi)
 492                                 panic("NMI indicates hardware failure");
 493                         break;
 494 #endif /* POWERFAIL_NMI */
 495 #endif /* DEV_ISA */
 496
 497                 case T_OFLOW:           /* integer overflow fault */
 498                         ucode = FPE_INTOVF;
 499                         i = SIGFPE;
 500                         break;
 501
 502                 case T_BOUND:           /* bounds check fault */
 503                         ucode = FPE_FLTSUB;
 504                         i = SIGFPE;
 505                         break;
 506
 507                 case T_DNA:
 508 #ifdef DEV_NPX
 509                         KASSERT(PCB_USER_FPU(td->td_pcb),
 510                             ("kernel FPU ctx has leaked"));
 511                         /* transparent fault (due to context switch "late") */
 512                         if (npxdna())
 513                                 goto userout;
 514 #endif
 515                         uprintf("pid %d killed due to lack of floating point\n",
 516                                 p->p_pid);
 517                         i = SIGKILL;
 518                         ucode = 0;
 519                         break;
 520
 521                 case T_FPOPFLT:         /* FPU operand fetch fault */
 522                         ucode = ILL_COPROC;
 523                         i = SIGILL;
 524                         break;
 525
 526                 case T_XMMFLT:          /* SIMD floating-point exception */
 527 #if defined(DEV_NPX) && !defined(CPU_DISABLE_SSE) && defined(I686_CPU)
 528                         ucode = npxtrap_sse();
 529                         if (ucode == -1)
 530                                 goto userout;
 531 #else
 532                         ucode = 0;
 533 #endif
 534                         i = SIGFPE;
 535                         break;
 536 #ifdef KDTRACE_HOOKS
 537                 case T_DTRACE_RET:
 538                         enable_intr();
 539                         fill_frame_regs(frame, &regs);
 540                         if (dtrace_return_probe_ptr != NULL &&
 541                             dtrace_return_probe_ptr(&regs) == 0)
 542                                 goto out;
 543                         break;
 544 #endif
 545                 }
 546         } else {
 547                 /* kernel trap */
 548
 549                 KASSERT(cold || td->td_ucred != NULL,
 550                     ("kernel trap doesn't have ucred"));
 551                 switch (type) {
 552                 case T_PAGEFLT:                 /* page fault */
 553                         (void) trap_pfault(frame, FALSE, eva);
 554                         goto out;
 555
 556                 case T_DNA:
 557 #ifdef DEV_NPX
 558                         KASSERT(!PCB_USER_FPU(td->td_pcb),
 559                             ("Unregistered use of FPU in kernel"));
 560                         if (npxdna())
 561                                 goto out;
 562 #endif
 563                         break;
 564
 565                 case T_ARITHTRAP:       /* arithmetic trap */
 566                 case T_XMMFLT:          /* SIMD floating-point exception */
 567                 case T_FPOPFLT:         /* FPU operand fetch fault */
 568                         /*
 569                          * XXXKIB for now disable any FPU traps in kernel
 570                          * handler registration seems to be overkill
 571                          */
 572                         trap_fatal(frame, 0);
 573                         goto out;
 574
 575                         /*
 576                          * The following two traps can happen in
 577                          * vm86 mode, and, if so, we want to handle
 578                          * them specially.
 579                          */
 580                 case T_PROTFLT:         /* general protection fault */
 581                 case T_STKFLT:          /* stack fault */
 582                         if (frame->tf_eflags & PSL_VM) {
 583                                 i = vm86_emulate((struct vm86frame *)frame);
 584                                 if (i != 0)
 585                                         /*
 586                                          * returns to original process
 587                                          */
 588                                         vm86_trap((struct vm86frame *)frame);
 589                                 goto out;
 590                         }
 591                         if (type == T_STKFLT)
 592                                 break;
 593
 594                         /* FALL THROUGH */
 595
 596                 case T_SEGNPFLT:        /* segment not present fault */
 597                         if (curpcb->pcb_flags & PCB_VM86CALL)
 598                                 break;
 599
 600                         /*
 601                          * Invalid %fs's and %gs's can be created using
 602                          * procfs or PT_SETREGS or by invalidating the
 603                          * underlying LDT entry.  This causes a fault
 604                          * in kernel mode when the kernel attempts to
 605                          * switch contexts.  Lose the bad context
 606                          * (XXX) so that we can continue, and generate
 607                          * a signal.
 608                          */
 609                         if (frame->tf_eip == (int)cpu_switch_load_gs) {
 610                                 curpcb->pcb_gs = 0;
 611 #if 0
 612                                 PROC_LOCK(p);
 613                                 kern_psignal(p, SIGBUS);
 614                                 PROC_UNLOCK(p);
 615 #endif
 616                                 goto out;
 617                         }
 618
 619                         if (td->td_intr_nesting_level != 0)
 620                                 break;
 621
 622                         /*
 623                          * Invalid segment selectors and out of bounds
 624                          * %eip's and %esp's can be set up in user mode.
 625                          * This causes a fault in kernel mode when the
 626                          * kernel tries to return to user mode.  We want
 627                          * to get this fault so that we can fix the
 628                          * problem here and not have to check all the
 629                          * selectors and pointers when the user changes
 630                          * them.
 631                          */
 632                         if (frame->tf_eip == (int)doreti_iret) {
 633                                 frame->tf_eip = (int)doreti_iret_fault;
 634                                 goto out;
 635                         }
 636                         if (frame->tf_eip == (int)doreti_popl_ds) {
 637                                 frame->tf_eip = (int)doreti_popl_ds_fault;
 638                                 goto out;
 639                         }
 640                         if (frame->tf_eip == (int)doreti_popl_es) {
 641                                 frame->tf_eip = (int)doreti_popl_es_fault;
 642                                 goto out;
 643                         }
 644                         if (frame->tf_eip == (int)doreti_popl_fs) {
 645                                 frame->tf_eip = (int)doreti_popl_fs_fault;
 646                                 goto out;
 647                         }
 648                         if (curpcb->pcb_onfault != NULL) {
 649                                 frame->tf_eip =
 650                                     (int)curpcb->pcb_onfault;
 651                                 goto out;
 652                         }
 653                         break;
 654
 655                 case T_TSSFLT:
 656                         /*
 657                          * PSL_NT can be set in user mode and isn't cleared
 658                          * automatically when the kernel is entered.  This
 659                          * causes a TSS fault when the kernel attempts to
 660                          * `iret' because the TSS link is uninitialized.  We
 661                          * want to get this fault so that we can fix the
 662                          * problem here and not every time the kernel is
 663                          * entered.
 664                          */
 665                         if (frame->tf_eflags & PSL_NT) {
 666                                 frame->tf_eflags &= ~PSL_NT;
 667                                 goto out;
 668                         }
 669                         break;
 670
 671                 case T_TRCTRAP:  /* trace trap */
 672                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall)) {
 673                                 /*
 674                                  * We've just entered system mode via the
 675                                  * syscall lcall.  Continue single stepping
 676                                  * silently until the syscall handler has
 677                                  * saved the flags.
 678                                  */
 679                                 goto out;
 680                         }
 681                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall) + 1) {
 682                                 /*
 683                                  * The syscall handler has now saved the
 684                                  * flags.  Stop single stepping it.
 685                                  */
 686                                 frame->tf_eflags &= ~PSL_T;
 687                                 goto out;
 688                         }
 689                         /*
 690                          * Ignore debug register trace traps due to
 691                          * accesses in the user's address space, which
 692                          * can happen under several conditions such as
 693                          * if a user sets a watchpoint on a buffer and
 694                          * then passes that buffer to a system call.
 695                          * We still want to get TRCTRAPS for addresses
 696                          * in kernel space because that is useful when
 697                          * debugging the kernel.
 698                          */
 699                         if (user_dbreg_trap() &&
 700                            !(curpcb->pcb_flags & PCB_VM86CALL)) {
 701                                 /*
 702                                  * Reset breakpoint bits because the
 703                                  * processor doesn't
 704                                  */
 705                                 load_dr6(rdr6() & 0xfffffff0);
 706                                 goto out;
 707                         }
 708                         /*
 709                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 710                          */
 711                 case T_BPTFLT:
 712                         /*
 713                          * If KDB is enabled, let it handle the debugger trap.
 714                          * Otherwise, debugger traps "can't happen".
 715                          */
 716 #ifdef KDB
 717                         if (kdb_trap(type, 0, frame))
 718                                 goto out;
 719 #endif
 720                         break;
 721
 722 #ifdef DEV_ISA
 723                 case T_NMI:
 724 #ifdef POWERFAIL_NMI
 725                         if (time_second - lastalert > 10) {
 726                                 log(LOG_WARNING, "NMI: power fail\n");
 727                                 sysbeep(880, hz);
 728                                 lastalert = time_second;
 729                         }
 730                         goto out;
 731 #else /* !POWERFAIL_NMI */
 732                         /* machine/parity/power fail/"kitchen sink" faults */
 733                         if (isa_nmi(code) == 0) {
 734 #ifdef KDB
 735                                 /*
 736                                  * NMI can be hooked up to a pushbutton
 737                                  * for debugging.
 738                                  */
 739                                 if (kdb_on_nmi) {
 740                                         printf ("NMI ... going to debugger\n");
 741                                         kdb_trap(type, 0, frame);
 742                                 }
 743 #endif /* KDB */
 744                                 goto out;
 745                         } else if (panic_on_nmi == 0)
 746                                 goto out;
 747                         /* FALLTHROUGH */
 748 #endif /* POWERFAIL_NMI */
 749 #endif /* DEV_ISA */
 750                 }
 751
 752                 trap_fatal(frame, eva);
 753                 goto out;
 754         }
 755
 756         /* Translate fault for emulators (e.g. Linux) */
 757         if (*p->p_sysent->sv_transtrap)
 758                 i = (*p->p_sysent->sv_transtrap)(i, type);
 759
 760         ksiginfo_init_trap(&ksi);
 761         ksi.ksi_signo = i;
 762         ksi.ksi_code = ucode;
 763         ksi.ksi_addr = (void *)addr;
 764         ksi.ksi_trapno = type;
 765         if (uprintf_signal) {
 766                 uprintf("pid %d comm %s: signal %d err %x code %d type %d "
 767                     "addr 0x%x esp 0x%08x eip 0x%08x "
 768                     "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
 769                     p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr,
 770                     frame->tf_esp, frame->tf_eip,
 771                     fubyte((void *)(frame->tf_eip + 0)),
 772                     fubyte((void *)(frame->tf_eip + 1)),
 773                     fubyte((void *)(frame->tf_eip + 2)),
 774                     fubyte((void *)(frame->tf_eip + 3)),
 775                     fubyte((void *)(frame->tf_eip + 4)),
 776                     fubyte((void *)(frame->tf_eip + 5)),
 777                     fubyte((void *)(frame->tf_eip + 6)),
 778                     fubyte((void *)(frame->tf_eip + 7)));
 779         }
 780         KASSERT((read_eflags() & PSL_I) != 0, ("interrupts disabled"));
 781         trapsignal(td, &ksi);
 782
 783 #ifdef DEBUG
 784         if (type <= MAX_TRAP_MSG) {
 785                 uprintf("fatal process exception: %s",
 786                         trap_msg[type]);
 787                 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
 788                         uprintf(", fault VA = 0x%lx", (u_long)eva);
 789                 uprintf("\n");
 790         }
 791 #endif
 792
 793 user:
 794         userret(td, frame);
 795         mtx_assert(&Giant, MA_NOTOWNED);
 796         KASSERT(PCB_USER_FPU(td->td_pcb),
 797             ("Return from trap with kernel FPU ctx leaked"));
 798 userout:
 799 out:
 800         return;
 801 }
 802
 803 static int
 804 trap_pfault(frame, usermode, eva)
 805         struct trapframe *frame;
 806         int usermode;
 807         vm_offset_t eva;
 808 {
 809         vm_offset_t va;
 810         struct vmspace *vm = NULL;
 811         vm_map_t map;
 812         int rv = 0;
 813         vm_prot_t ftype;
 814         struct thread *td = curthread;
 815         struct proc *p = td->td_proc;
 816
 817         if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
 818                 /*
 819                  * Due to both processor errata and lazy TLB invalidation when
 820                  * access restrictions are removed from virtual pages, memory
 821                  * accesses that are allowed by the physical mapping layer may
 822                  * nonetheless cause one spurious page fault per virtual page.
 823                  * When the thread is executing a "no faulting" section that
 824                  * is bracketed by vm_fault_{disable,enable}_pagefaults(),
 825                  * every page fault is treated as a spurious page fault,
 826                  * unless it accesses the same virtual address as the most
 827                  * recent page fault within the same "no faulting" section.
 828                  */
 829                 if (td->td_md.md_spurflt_addr != eva ||
 830                     (td->td_pflags & TDP_RESETSPUR) != 0) {
 831                         /*
 832                          * Do nothing to the TLB.  A stale TLB entry is
 833                          * flushed automatically by a page fault.
 834                          */
 835                         td->td_md.md_spurflt_addr = eva;
 836                         td->td_pflags &= ~TDP_RESETSPUR;
 837                         return (0);
 838                 }
 839         } else {
 840                 /*
 841                  * If we get a page fault while in a critical section, then
 842                  * it is most likely a fatal kernel page fault.  The kernel
 843                  * is already going to panic trying to get a sleep lock to
 844                  * do the VM lookup, so just consider it a fatal trap so the
 845                  * kernel can print out a useful trap message and even get
 846                  * to the debugger.
 847                  *
 848                  * If we get a page fault while holding a non-sleepable
 849                  * lock, then it is most likely a fatal kernel page fault.
 850                  * If WITNESS is enabled, then it's going to whine about
 851                  * bogus LORs with various VM locks, so just skip to the
 852                  * fatal trap handling directly.
 853                  */
 854                 if (td->td_critnest != 0 ||
 855                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 856                     "Kernel page fault") != 0) {
 857                         trap_fatal(frame, eva);
 858                         return (-1);
 859                 }
 860         }
 861         va = trunc_page(eva);
 862         if (va >= KERNBASE) {
 863                 /*
 864                  * Don't allow user-mode faults in kernel address space.
 865                  * An exception:  if the faulting address is the invalid
 866                  * instruction entry in the IDT, then the Intel Pentium
 867                  * F00F bug workaround was triggered, and we need to
 868                  * treat it is as an illegal instruction, and not a page
 869                  * fault.
 870                  */
 871 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 872                 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
 873                         return -2;
 874 #endif
 875                 if (usermode)
 876                         goto nogo;
 877
 878                 map = kernel_map;
 879         } else {
 880                 /*
 881                  * This is a fault on non-kernel virtual memory.
 882                  * vm is initialized above to NULL. If curproc is NULL
 883                  * or curproc->p_vmspace is NULL the fault is fatal.
 884                  */
 885                 if (p != NULL)
 886                         vm = p->p_vmspace;
 887
 888                 if (vm == NULL)
 889                         goto nogo;
 890
 891                 map = &vm->vm_map;
 892                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 893                     curpcb->pcb_onfault == NULL)) {
 894                         trap_fatal(frame, eva);
 895                         return (-1);
 896                 }
 897         }
 898
 899         /*
 900          * PGEX_I is defined only if the execute disable bit capability is
 901          * supported and enabled.
 902          */
 903         if (frame->tf_err & PGEX_W)
 904                 ftype = VM_PROT_WRITE;
 905 #ifdef PAE
 906         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 907                 ftype = VM_PROT_EXECUTE;
 908 #endif
 909         else
 910                 ftype = VM_PROT_READ;
 911
 912         if (map != kernel_map) {
 913                 /*
 914                  * Keep swapout from messing with us during this
 915                  *      critical time.
 916                  */
 917                 PROC_LOCK(p);
 918                 ++p->p_lock;
 919                 PROC_UNLOCK(p);
 920
 921                 /* Fault in the user page: */
 922                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 923
 924                 PROC_LOCK(p);
 925                 --p->p_lock;
 926                 PROC_UNLOCK(p);
 927         } else {
 928                 /*
 929                  * Don't have to worry about process locking or stacks in the
 930                  * kernel.
 931                  */
 932                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 933         }
 934         if (rv == KERN_SUCCESS) {
 935 #ifdef HWPMC_HOOKS
 936                 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
 937                         PMC_SOFT_CALL_TF( , , page_fault, all, frame);
 938                         if (ftype == VM_PROT_READ)
 939                                 PMC_SOFT_CALL_TF( , , page_fault, read,
 940                                     frame);
 941                         else
 942                                 PMC_SOFT_CALL_TF( , , page_fault, write,
 943                                     frame);
 944                 }
 945 #endif
 946                 return (0);
 947         }
 948 nogo:
 949         if (!usermode) {
 950                 if (td->td_intr_nesting_level == 0 &&
 951                     curpcb->pcb_onfault != NULL) {
 952                         frame->tf_eip = (int)curpcb->pcb_onfault;
 953                         return (0);
 954                 }
 955                 trap_fatal(frame, eva);
 956                 return (-1);
 957         }
 958
 959         return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 960 }
 961
 962 static void
 963 trap_fatal(frame, eva)
 964         struct trapframe *frame;
 965         vm_offset_t eva;
 966 {
 967         int code, ss, esp;
 968         u_int type;
 969         struct soft_segment_descriptor softseg;
 970         char *msg;
 971
 972         code = frame->tf_err;
 973         type = frame->tf_trapno;
 974         sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
 975
 976         if (type <= MAX_TRAP_MSG)
 977                 msg = trap_msg[type];
 978         else
 979                 msg = "UNKNOWN";
 980         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 981             frame->tf_eflags & PSL_VM ? "vm86" :
 982             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
 983 #ifdef SMP
 984         /* two separate prints in case of a trap on an unmapped page */
 985         printf("cpuid = %d; ", PCPU_GET(cpuid));
 986         printf("apic id = %02x\n", PCPU_GET(apic_id));
 987 #endif
 988         if (type == T_PAGEFLT) {
 989                 printf("fault virtual address   = 0x%x\n", eva);
 990                 printf("fault code              = %s %s, %s\n",
 991                         code & PGEX_U ? "user" : "supervisor",
 992                         code & PGEX_W ? "write" : "read",
 993                         code & PGEX_P ? "protection violation" : "page not present");
 994         }
 995         printf("instruction pointer     = 0x%x:0x%x\n",
 996                frame->tf_cs & 0xffff, frame->tf_eip);
 997         if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
 998                 ss = frame->tf_ss & 0xffff;
 999                 esp = frame->tf_esp;
1000         } else {
1001                 ss = GSEL(GDATA_SEL, SEL_KPL);
1002                 esp = (int)&frame->tf_esp;
1003         }
1004         printf("stack pointer           = 0x%x:0x%x\n", ss, esp);
1005         printf("frame pointer           = 0x%x:0x%x\n", ss, frame->tf_ebp);
1006         printf("code segment            = base 0x%x, limit 0x%x, type 0x%x\n",
1007                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1008         printf("                        = DPL %d, pres %d, def32 %d, gran %d\n",
1009                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1010                softseg.ssd_gran);
1011         printf("processor eflags        = ");
1012         if (frame->tf_eflags & PSL_T)
1013                 printf("trace trap, ");
1014         if (frame->tf_eflags & PSL_I)
1015                 printf("interrupt enabled, ");
1016         if (frame->tf_eflags & PSL_NT)
1017                 printf("nested task, ");
1018         if (frame->tf_eflags & PSL_RF)
1019                 printf("resume, ");
1020         if (frame->tf_eflags & PSL_VM)
1021                 printf("vm86, ");
1022         printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1023         printf("current process         = %d (%s)\n",
1024             curproc->p_pid, curthread->td_name);
1025
1026 #ifdef KDB
1027         if (debugger_on_panic || kdb_active) {
1028                 frame->tf_err = eva;    /* smuggle fault address to ddb */
1029                 if (kdb_trap(type, 0, frame)) {
1030                         frame->tf_err = code;   /* restore error code */
1031                         return;
1032                 }
1033                 frame->tf_err = code;           /* restore error code */
1034         }
1035 #endif
1036         printf("trap number             = %d\n", type);
1037         if (type <= MAX_TRAP_MSG)
1038                 panic("%s", trap_msg[type]);
1039         else
1040                 panic("unknown/reserved trap");
1041 }
1042
1043 /*
1044  * Double fault handler. Called when a fault occurs while writing
1045  * a frame for a trap/exception onto the stack. This usually occurs
1046  * when the stack overflows (such is the case with infinite recursion,
1047  * for example).
1048  *
1049  * XXX Note that the current PTD gets replaced by IdlePTD when the
1050  * task switch occurs. This means that the stack that was active at
1051  * the time of the double fault is not available at <kstack> unless
1052  * the machine was idle when the double fault occurred. The downside
1053  * of this is that "trace <ebp>" in ddb won't work.
1054  */
1055 void
1056 dblfault_handler()
1057 {
1058 #ifdef KDTRACE_HOOKS
1059         if (dtrace_doubletrap_func != NULL)
1060                 (*dtrace_doubletrap_func)();
1061 #endif
1062         printf("\nFatal double fault:\n");
1063         printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip));
1064         printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp));
1065         printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp));
1066 #ifdef SMP
1067         /* two separate prints in case of a trap on an unmapped page */
1068         printf("cpuid = %d; ", PCPU_GET(cpuid));
1069         printf("apic id = %02x\n", PCPU_GET(apic_id));
1070 #endif
1071         panic("double fault");
1072 }
1073
1074 int
1075 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
1076 {
1077         struct proc *p;
1078         struct trapframe *frame;
1079         caddr_t params;
1080         int error;
1081
1082         p = td->td_proc;
1083         frame = td->td_frame;
1084
1085         params = (caddr_t)frame->tf_esp + sizeof(int);
1086         sa->code = frame->tf_eax;
1087
1088         /*
1089          * Need to check if this is a 32 bit or 64 bit syscall.
1090          */
1091         if (sa->code == SYS_syscall) {
1092                 /*
1093                  * Code is first argument, followed by actual args.
1094                  */
1095                 sa->code = fuword(params);
1096                 params += sizeof(int);
1097         } else if (sa->code == SYS___syscall) {
1098                 /*
1099                  * Like syscall, but code is a quad, so as to maintain
1100                  * quad alignment for the rest of the arguments.
1101                  */
1102                 sa->code = fuword(params);
1103                 params += sizeof(quad_t);
1104         }
1105
1106         if (p->p_sysent->sv_mask)
1107                 sa->code &= p->p_sysent->sv_mask;
1108         if (sa->code >= p->p_sysent->sv_size)
1109                 sa->callp = &p->p_sysent->sv_table[0];
1110         else
1111                 sa->callp = &p->p_sysent->sv_table[sa->code];
1112         sa->narg = sa->callp->sy_narg;
1113
1114         if (params != NULL && sa->narg != 0)
1115                 error = copyin(params, (caddr_t)sa->args,
1116                     (u_int)(sa->narg * sizeof(int)));
1117         else
1118                 error = 0;
1119
1120         if (error == 0) {
1121                 td->td_retval[0] = 0;
1122                 td->td_retval[1] = frame->tf_edx;
1123         }
1124
1125         return (error);
1126 }
1127
1128 #include "../../kern/subr_syscall.c"
1129
1130 /*
1131  * syscall - system call request C handler.  A system call is
1132  * essentially treated as a trap by reusing the frame layout.
1133  */
1134 void
1135 syscall(struct trapframe *frame)
1136 {
1137         struct thread *td;
1138         struct syscall_args sa;
1139         register_t orig_tf_eflags;
1140         int error;
1141         ksiginfo_t ksi;
1142
1143 #ifdef DIAGNOSTIC
1144         if (ISPL(frame->tf_cs) != SEL_UPL) {
1145                 panic("syscall");
1146                 /* NOT REACHED */
1147         }
1148 #endif
1149         orig_tf_eflags = frame->tf_eflags;
1150
1151         td = curthread;
1152         td->td_frame = frame;
1153
1154         error = syscallenter(td, &sa);
1155
1156         /*
1157          * Traced syscall.
1158          */
1159         if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1160                 frame->tf_eflags &= ~PSL_T;
1161                 ksiginfo_init_trap(&ksi);
1162                 ksi.ksi_signo = SIGTRAP;
1163                 ksi.ksi_code = TRAP_TRACE;
1164                 ksi.ksi_addr = (void *)frame->tf_eip;
1165                 trapsignal(td, &ksi);
1166         }
1167
1168         KASSERT(PCB_USER_FPU(td->td_pcb),
1169             ("System call %s returning with kernel FPU ctx leaked",
1170              syscallname(td->td_proc, sa.code)));
1171         KASSERT(td->td_pcb->pcb_save == &td->td_pcb->pcb_user_save,
1172             ("System call %s returning with mangled pcb_save",
1173              syscallname(td->td_proc, sa.code)));
1174
1175         syscallret(td, error, &sa);
1176 }