sys/amd64/amd64/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  * AMD64 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
  54 #include <sys/param.h>
  55 #include <sys/bus.h>
  56 #include <sys/systm.h>
  57 #include <sys/proc.h>
  58 #include <sys/pioctl.h>
  59 #include <sys/ptrace.h>
  60 #include <sys/kdb.h>
  61 #include <sys/kernel.h>
  62 #include <sys/ktr.h>
  63 #include <sys/lock.h>
  64 #include <sys/mutex.h>
  65 #include <sys/resourcevar.h>
  66 #include <sys/signalvar.h>
  67 #include <sys/syscall.h>
  68 #include <sys/sysctl.h>
  69 #include <sys/sysent.h>
  70 #include <sys/uio.h>
  71 #include <sys/vmmeter.h>
  72 #ifdef HWPMC_HOOKS
  73 #include <sys/pmckern.h>
  74 #endif
  75
  76 #include <vm/vm.h>
  77 #include <vm/vm_param.h>
  78 #include <vm/pmap.h>
  79 #include <vm/vm_kern.h>
  80 #include <vm/vm_map.h>
  81 #include <vm/vm_page.h>
  82 #include <vm/vm_extern.h>
  83
  84 #include <machine/cpu.h>
  85 #include <machine/intr_machdep.h>
  86 #include <x86/mca.h>
  87 #include <machine/md_var.h>
  88 #include <machine/pcb.h>
  89 #ifdef SMP
  90 #include <machine/smp.h>
  91 #endif
  92 #include <machine/tss.h>
  93
  94 #ifdef KDTRACE_HOOKS
  95 #include <sys/dtrace_bsd.h>
  96
  97 /*
  98  * This is a hook which is initialised by the dtrace module
  99  * to handle traps which might occur during DTrace probe
 100  * execution.
 101  */
 102 dtrace_trap_func_t      dtrace_trap_func;
 103
 104 dtrace_doubletrap_func_t        dtrace_doubletrap_func;
 105
 106 /*
 107  * This is a hook which is initialised by the systrace module
 108  * when it is loaded. This keeps the DTrace syscall provider
 109  * implementation opaque.
 110  */
 111 systrace_probe_func_t   systrace_probe_func;
 112
 113 /*
 114  * These hooks are necessary for the pid, usdt and fasttrap providers.
 115  */
 116 dtrace_fasttrap_probe_ptr_t     dtrace_fasttrap_probe_ptr;
 117 dtrace_pid_probe_ptr_t          dtrace_pid_probe_ptr;
 118 dtrace_return_probe_ptr_t       dtrace_return_probe_ptr;
 119 #endif
 120
 121 extern void trap(struct trapframe *frame);
 122 extern void syscall(struct trapframe *frame);
 123 void dblfault_handler(struct trapframe *frame);
 124
 125 static int trap_pfault(struct trapframe *, int);
 126 static void trap_fatal(struct trapframe *, vm_offset_t);
 127
 128 #define MAX_TRAP_MSG            30
 129 static char *trap_msg[] = {
 130         "",                                     /*  0 unused */
 131         "privileged instruction fault",         /*  1 T_PRIVINFLT */
 132         "",                                     /*  2 unused */
 133         "breakpoint instruction fault",         /*  3 T_BPTFLT */
 134         "",                                     /*  4 unused */
 135         "",                                     /*  5 unused */
 136         "arithmetic trap",                      /*  6 T_ARITHTRAP */
 137         "",                                     /*  7 unused */
 138         "",                                     /*  8 unused */
 139         "general protection fault",             /*  9 T_PROTFLT */
 140         "trace trap",                           /* 10 T_TRCTRAP */
 141         "",                                     /* 11 unused */
 142         "page fault",                           /* 12 T_PAGEFLT */
 143         "",                                     /* 13 unused */
 144         "alignment fault",                      /* 14 T_ALIGNFLT */
 145         "",                                     /* 15 unused */
 146         "",                                     /* 16 unused */
 147         "",                                     /* 17 unused */
 148         "integer divide fault",                 /* 18 T_DIVIDE */
 149         "non-maskable interrupt trap",          /* 19 T_NMI */
 150         "overflow trap",                        /* 20 T_OFLOW */
 151         "FPU bounds check fault",               /* 21 T_BOUND */
 152         "FPU device not available",             /* 22 T_DNA */
 153         "double fault",                         /* 23 T_DOUBLEFLT */
 154         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
 155         "invalid TSS fault",                    /* 25 T_TSSFLT */
 156         "segment not present fault",            /* 26 T_SEGNPFLT */
 157         "stack fault",                          /* 27 T_STKFLT */
 158         "machine check trap",                   /* 28 T_MCHK */
 159         "SIMD floating-point exception",        /* 29 T_XMMFLT */
 160         "reserved (unknown) fault",             /* 30 T_RESERVED */
 161 };
 162
 163 #ifdef KDB
 164 static int kdb_on_nmi = 1;
 165 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
 166         &kdb_on_nmi, 0, "Go to KDB on NMI");
 167 TUNABLE_INT("machdep.kdb_on_nmi", &kdb_on_nmi);
 168 #endif
 169 static int panic_on_nmi = 1;
 170 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
 171         &panic_on_nmi, 0, "Panic on NMI");
 172 TUNABLE_INT("machdep.panic_on_nmi", &panic_on_nmi);
 173 static int prot_fault_translation = 0;
 174 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
 175         &prot_fault_translation, 0, "Select signal to deliver on protection fault");
 176
 177 /*
 178  * Exception, fault, and trap interface to the FreeBSD kernel.
 179  * This common code is called from assembly language IDT gate entry
 180  * routines that prepare a suitable stack frame, and restore this
 181  * frame after the exception has been processed.
 182  */
 183
 184 void
 185 trap(struct trapframe *frame)
 186 {
 187         struct thread *td = curthread;
 188         struct proc *p = td->td_proc;
 189         int i = 0, ucode = 0, code;
 190         u_int type;
 191         register_t addr = 0;
 192         ksiginfo_t ksi;
 193
 194         PCPU_INC(cnt.v_trap);
 195         type = frame->tf_trapno;
 196
 197 #ifdef SMP
 198         /* Handler for NMI IPIs used for stopping CPUs. */
 199         if (type == T_NMI) {
 200                  if (ipi_nmi_handler() == 0)
 201                            goto out;
 202         }
 203 #endif /* SMP */
 204
 205 #ifdef KDB
 206         if (kdb_active) {
 207                 kdb_reenter();
 208                 goto out;
 209         }
 210 #endif
 211
 212         if (type == T_RESERVED) {
 213                 trap_fatal(frame, 0);
 214                 goto out;
 215         }
 216
 217 #ifdef  HWPMC_HOOKS
 218         /*
 219          * CPU PMCs interrupt using an NMI.  If the PMC module is
 220          * active, pass the 'rip' value to the PMC module's interrupt
 221          * handler.  A return value of '1' from the handler means that
 222          * the NMI was handled by it and we can return immediately.
 223          */
 224         if (type == T_NMI && pmc_intr &&
 225             (*pmc_intr)(PCPU_GET(cpuid), frame))
 226                 goto out;
 227 #endif
 228
 229         if (type == T_MCHK) {
 230                 if (!mca_intr())
 231                         trap_fatal(frame, 0);
 232                 goto out;
 233         }
 234
 235 #ifdef KDTRACE_HOOKS
 236         /*
 237          * A trap can occur while DTrace executes a probe. Before
 238          * executing the probe, DTrace blocks re-scheduling and sets
 239          * a flag in it's per-cpu flags to indicate that it doesn't
 240          * want to fault. On returning from the probe, the no-fault
 241          * flag is cleared and finally re-scheduling is enabled.
 242          *
 243          * If the DTrace kernel module has registered a trap handler,
 244          * call it and if it returns non-zero, assume that it has
 245          * handled the trap and modified the trap frame so that this
 246          * function can return normally.
 247          */
 248         if (dtrace_trap_func != NULL)
 249                 if ((*dtrace_trap_func)(frame, type))
 250                         goto out;
 251         if (type == T_DTRACE_PROBE || type == T_DTRACE_RET ||
 252             type == T_BPTFLT) {
 253                 struct reg regs;
 254
 255                 fill_frame_regs(frame, &regs);
 256                 if (type == T_DTRACE_PROBE &&
 257                     dtrace_fasttrap_probe_ptr != NULL &&
 258                     dtrace_fasttrap_probe_ptr(&regs) == 0)
 259                                 goto out;
 260                 if (type == T_BPTFLT &&
 261                     dtrace_pid_probe_ptr != NULL &&
 262                     dtrace_pid_probe_ptr(&regs) == 0)
 263                                 goto out;
 264                 if (type == T_DTRACE_RET &&
 265                     dtrace_return_probe_ptr != NULL &&
 266                     dtrace_return_probe_ptr(&regs) == 0)
 267                         goto out;
 268
 269         }
 270 #endif
 271
 272         if ((frame->tf_rflags & PSL_I) == 0) {
 273                 /*
 274                  * Buggy application or kernel code has disabled
 275                  * interrupts and then trapped.  Enabling interrupts
 276                  * now is wrong, but it is better than running with
 277                  * interrupts disabled until they are accidentally
 278                  * enabled later.
 279                  */
 280                 if (ISPL(frame->tf_cs) == SEL_UPL)
 281                         uprintf(
 282                             "pid %ld (%s): trap %d with interrupts disabled\n",
 283                             (long)curproc->p_pid, curthread->td_name, type);
 284                 else if (type != T_NMI && type != T_BPTFLT &&
 285                     type != T_TRCTRAP) {
 286                         /*
 287                          * XXX not quite right, since this may be for a
 288                          * multiple fault in user mode.
 289                          */
 290                         printf("kernel trap %d with interrupts disabled\n",
 291                             type);
 292
 293                         /*
 294                          * We shouldn't enable interrupts while holding a
 295                          * spin lock.
 296                          */
 297                         if (td->td_md.md_spinlock_count == 0)
 298                                 enable_intr();
 299                 }
 300         }
 301
 302         code = frame->tf_err;
 303         if (type == T_PAGEFLT) {
 304                 /*
 305                  * If we get a page fault while in a critical section, then
 306                  * it is most likely a fatal kernel page fault.  The kernel
 307                  * is already going to panic trying to get a sleep lock to
 308                  * do the VM lookup, so just consider it a fatal trap so the
 309                  * kernel can print out a useful trap message and even get
 310                  * to the debugger.
 311                  *
 312                  * If we get a page fault while holding a non-sleepable
 313                  * lock, then it is most likely a fatal kernel page fault.
 314                  * If WITNESS is enabled, then it's going to whine about
 315                  * bogus LORs with various VM locks, so just skip to the
 316                  * fatal trap handling directly.
 317                  */
 318                 if (td->td_critnest != 0 ||
 319                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 320                     "Kernel page fault") != 0)
 321                         trap_fatal(frame, frame->tf_addr);
 322         }
 323
 324         if (ISPL(frame->tf_cs) == SEL_UPL) {
 325                 /* user trap */
 326
 327                 td->td_pticks = 0;
 328                 td->td_frame = frame;
 329                 addr = frame->tf_rip;
 330                 if (td->td_ucred != p->p_ucred)
 331                         cred_update_thread(td);
 332
 333                 switch (type) {
 334                 case T_PRIVINFLT:       /* privileged instruction fault */
 335                         i = SIGILL;
 336                         ucode = ILL_PRVOPC;
 337                         break;
 338
 339                 case T_BPTFLT:          /* bpt instruction fault */
 340                 case T_TRCTRAP:         /* trace trap */
 341                         enable_intr();
 342                         frame->tf_rflags &= ~PSL_T;
 343                         i = SIGTRAP;
 344                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 345                         break;
 346
 347                 case T_ARITHTRAP:       /* arithmetic trap */
 348                         ucode = fputrap();
 349                         if (ucode == -1)
 350                                 goto userout;
 351                         i = SIGFPE;
 352                         break;
 353
 354                 case T_PROTFLT:         /* general protection fault */
 355                         i = SIGBUS;
 356                         ucode = BUS_OBJERR;
 357                         break;
 358                 case T_STKFLT:          /* stack fault */
 359                 case T_SEGNPFLT:        /* segment not present fault */
 360                         i = SIGBUS;
 361                         ucode = BUS_ADRERR;
 362                         break;
 363                 case T_TSSFLT:          /* invalid TSS fault */
 364                         i = SIGBUS;
 365                         ucode = BUS_OBJERR;
 366                         break;
 367                 case T_DOUBLEFLT:       /* double fault */
 368                 default:
 369                         i = SIGBUS;
 370                         ucode = BUS_OBJERR;
 371                         break;
 372
 373                 case T_PAGEFLT:         /* page fault */
 374                         addr = frame->tf_addr;
 375                         i = trap_pfault(frame, TRUE);
 376                         if (i == -1)
 377                                 goto userout;
 378                         if (i == 0)
 379                                 goto user;
 380
 381                         if (i == SIGSEGV)
 382                                 ucode = SEGV_MAPERR;
 383                         else {
 384                                 if (prot_fault_translation == 0) {
 385                                         /*
 386                                          * Autodetect.
 387                                          * This check also covers the images
 388                                          * without the ABI-tag ELF note.
 389                                          */
 390                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 391                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 392                                                 i = SIGSEGV;
 393                                                 ucode = SEGV_ACCERR;
 394                                         } else {
 395                                                 i = SIGBUS;
 396                                                 ucode = BUS_PAGE_FAULT;
 397                                         }
 398                                 } else if (prot_fault_translation == 1) {
 399                                         /*
 400                                          * Always compat mode.
 401                                          */
 402                                         i = SIGBUS;
 403                                         ucode = BUS_PAGE_FAULT;
 404                                 } else {
 405                                         /*
 406                                          * Always SIGSEGV mode.
 407                                          */
 408                                         i = SIGSEGV;
 409                                         ucode = SEGV_ACCERR;
 410                                 }
 411                         }
 412                         break;
 413
 414                 case T_DIVIDE:          /* integer divide fault */
 415                         ucode = FPE_INTDIV;
 416                         i = SIGFPE;
 417                         break;
 418
 419 #ifdef DEV_ISA
 420                 case T_NMI:
 421                         /* machine/parity/power fail/"kitchen sink" faults */
 422                         if (isa_nmi(code) == 0) {
 423 #ifdef KDB
 424                                 /*
 425                                  * NMI can be hooked up to a pushbutton
 426                                  * for debugging.
 427                                  */
 428                                 if (kdb_on_nmi) {
 429                                         printf ("NMI ... going to debugger\n");
 430                                         kdb_trap(type, 0, frame);
 431                                 }
 432 #endif /* KDB */
 433                                 goto userout;
 434                         } else if (panic_on_nmi)
 435                                 panic("NMI indicates hardware failure");
 436                         break;
 437 #endif /* DEV_ISA */
 438
 439                 case T_OFLOW:           /* integer overflow fault */
 440                         ucode = FPE_INTOVF;
 441                         i = SIGFPE;
 442                         break;
 443
 444                 case T_BOUND:           /* bounds check fault */
 445                         ucode = FPE_FLTSUB;
 446                         i = SIGFPE;
 447                         break;
 448
 449                 case T_DNA:
 450                         /* transparent fault (due to context switch "late") */
 451                         KASSERT(PCB_USER_FPU(td->td_pcb),
 452                             ("kernel FPU ctx has leaked"));
 453                         fpudna();
 454                         goto userout;
 455
 456                 case T_FPOPFLT:         /* FPU operand fetch fault */
 457                         ucode = ILL_COPROC;
 458                         i = SIGILL;
 459                         break;
 460
 461                 case T_XMMFLT:          /* SIMD floating-point exception */
 462                         ucode = 0; /* XXX */
 463                         i = SIGFPE;
 464                         break;
 465                 }
 466         } else {
 467                 /* kernel trap */
 468
 469                 KASSERT(cold || td->td_ucred != NULL,
 470                     ("kernel trap doesn't have ucred"));
 471                 switch (type) {
 472                 case T_PAGEFLT:                 /* page fault */
 473                         (void) trap_pfault(frame, FALSE);
 474                         goto out;
 475
 476                 case T_DNA:
 477                         KASSERT(!PCB_USER_FPU(td->td_pcb),
 478                             ("Unregistered use of FPU in kernel"));
 479                         fpudna();
 480                         goto out;
 481
 482                 case T_ARITHTRAP:       /* arithmetic trap */
 483                 case T_XMMFLT:          /* SIMD floating-point exception */
 484                 case T_FPOPFLT:         /* FPU operand fetch fault */
 485                         /*
 486                          * XXXKIB for now disable any FPU traps in kernel
 487                          * handler registration seems to be overkill
 488                          */
 489                         trap_fatal(frame, 0);
 490                         goto out;
 491
 492                 case T_STKFLT:          /* stack fault */
 493                         break;
 494
 495                 case T_PROTFLT:         /* general protection fault */
 496                 case T_SEGNPFLT:        /* segment not present fault */
 497                         if (td->td_intr_nesting_level != 0)
 498                                 break;
 499
 500                         /*
 501                          * Invalid segment selectors and out of bounds
 502                          * %rip's and %rsp's can be set up in user mode.
 503                          * This causes a fault in kernel mode when the
 504                          * kernel tries to return to user mode.  We want
 505                          * to get this fault so that we can fix the
 506                          * problem here and not have to check all the
 507                          * selectors and pointers when the user changes
 508                          * them.
 509                          */
 510                         if (frame->tf_rip == (long)doreti_iret) {
 511                                 frame->tf_rip = (long)doreti_iret_fault;
 512                                 goto out;
 513                         }
 514                         if (frame->tf_rip == (long)ld_ds) {
 515                                 frame->tf_rip = (long)ds_load_fault;
 516                                 goto out;
 517                         }
 518                         if (frame->tf_rip == (long)ld_es) {
 519                                 frame->tf_rip = (long)es_load_fault;
 520                                 goto out;
 521                         }
 522                         if (frame->tf_rip == (long)ld_fs) {
 523                                 frame->tf_rip = (long)fs_load_fault;
 524                                 goto out;
 525                         }
 526                         if (frame->tf_rip == (long)ld_gs) {
 527                                 frame->tf_rip = (long)gs_load_fault;
 528                                 goto out;
 529                         }
 530                         if (frame->tf_rip == (long)ld_gsbase) {
 531                                 frame->tf_rip = (long)gsbase_load_fault;
 532                                 goto out;
 533                         }
 534                         if (frame->tf_rip == (long)ld_fsbase) {
 535                                 frame->tf_rip = (long)fsbase_load_fault;
 536                                 goto out;
 537                         }
 538                         if (PCPU_GET(curpcb)->pcb_onfault != NULL) {
 539                                 frame->tf_rip =
 540                                     (long)PCPU_GET(curpcb)->pcb_onfault;
 541                                 goto out;
 542                         }
 543                         break;
 544
 545                 case T_TSSFLT:
 546                         /*
 547                          * PSL_NT can be set in user mode and isn't cleared
 548                          * automatically when the kernel is entered.  This
 549                          * causes a TSS fault when the kernel attempts to
 550                          * `iret' because the TSS link is uninitialized.  We
 551                          * want to get this fault so that we can fix the
 552                          * problem here and not every time the kernel is
 553                          * entered.
 554                          */
 555                         if (frame->tf_rflags & PSL_NT) {
 556                                 frame->tf_rflags &= ~PSL_NT;
 557                                 goto out;
 558                         }
 559                         break;
 560
 561                 case T_TRCTRAP:  /* trace trap */
 562                         /*
 563                          * Ignore debug register trace traps due to
 564                          * accesses in the user's address space, which
 565                          * can happen under several conditions such as
 566                          * if a user sets a watchpoint on a buffer and
 567                          * then passes that buffer to a system call.
 568                          * We still want to get TRCTRAPS for addresses
 569                          * in kernel space because that is useful when
 570                          * debugging the kernel.
 571                          */
 572                         if (user_dbreg_trap()) {
 573                                 /*
 574                                  * Reset breakpoint bits because the
 575                                  * processor doesn't
 576                                  */
 577                                 /* XXX check upper bits here */
 578                                 load_dr6(rdr6() & 0xfffffff0);
 579                                 goto out;
 580                         }
 581                         /*
 582                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 583                          */
 584                 case T_BPTFLT:
 585                         /*
 586                          * If KDB is enabled, let it handle the debugger trap.
 587                          * Otherwise, debugger traps "can't happen".
 588                          */
 589 #ifdef KDB
 590                         if (kdb_trap(type, 0, frame))
 591                                 goto out;
 592 #endif
 593                         break;
 594
 595 #ifdef DEV_ISA
 596                 case T_NMI:
 597                         /* machine/parity/power fail/"kitchen sink" faults */
 598                         if (isa_nmi(code) == 0) {
 599 #ifdef KDB
 600                                 /*
 601                                  * NMI can be hooked up to a pushbutton
 602                                  * for debugging.
 603                                  */
 604                                 if (kdb_on_nmi) {
 605                                         printf ("NMI ... going to debugger\n");
 606                                         kdb_trap(type, 0, frame);
 607                                 }
 608 #endif /* KDB */
 609                                 goto out;
 610                         } else if (panic_on_nmi == 0)
 611                                 goto out;
 612                         /* FALLTHROUGH */
 613 #endif /* DEV_ISA */
 614                 }
 615
 616                 trap_fatal(frame, 0);
 617                 goto out;
 618         }
 619
 620         /* Translate fault for emulators (e.g. Linux) */
 621         if (*p->p_sysent->sv_transtrap)
 622                 i = (*p->p_sysent->sv_transtrap)(i, type);
 623
 624         ksiginfo_init_trap(&ksi);
 625         ksi.ksi_signo = i;
 626         ksi.ksi_code = ucode;
 627         ksi.ksi_trapno = type;
 628         ksi.ksi_addr = (void *)addr;
 629         trapsignal(td, &ksi);
 630
 631 user:
 632         userret(td, frame);
 633         mtx_assert(&Giant, MA_NOTOWNED);
 634         KASSERT(PCB_USER_FPU(td->td_pcb),
 635             ("Return from trap with kernel FPU ctx leaked"));
 636 userout:
 637 out:
 638         return;
 639 }
 640
 641 static int
 642 trap_pfault(frame, usermode)
 643         struct trapframe *frame;
 644         int usermode;
 645 {
 646         vm_offset_t va;
 647         struct vmspace *vm = NULL;
 648         vm_map_t map;
 649         int rv = 0;
 650         vm_prot_t ftype;
 651         struct thread *td = curthread;
 652         struct proc *p = td->td_proc;
 653         vm_offset_t eva = frame->tf_addr;
 654
 655         va = trunc_page(eva);
 656         if (va >= VM_MIN_KERNEL_ADDRESS) {
 657                 /*
 658                  * Don't allow user-mode faults in kernel address space.
 659                  */
 660                 if (usermode)
 661                         goto nogo;
 662
 663                 map = kernel_map;
 664         } else {
 665                 /*
 666                  * This is a fault on non-kernel virtual memory.
 667                  * vm is initialized above to NULL. If curproc is NULL
 668                  * or curproc->p_vmspace is NULL the fault is fatal.
 669                  */
 670                 if (p != NULL)
 671                         vm = p->p_vmspace;
 672
 673                 if (vm == NULL)
 674                         goto nogo;
 675
 676                 map = &vm->vm_map;
 677
 678                 /*
 679                  * When accessing a usermode address, kernel must be
 680                  * ready to accept the page fault, and provide a
 681                  * handling routine.  Since accessing the address
 682                  * without the handler is a bug, do not try to handle
 683                  * it normally, and panic immediately.
 684                  */
 685                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 686                     PCPU_GET(curpcb)->pcb_onfault == NULL)) {
 687                         trap_fatal(frame, eva);
 688                         return (-1);
 689                 }
 690         }
 691
 692         /*
 693          * PGEX_I is defined only if the execute disable bit capability is
 694          * supported and enabled.
 695          */
 696         if (frame->tf_err & PGEX_W)
 697                 ftype = VM_PROT_WRITE;
 698         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 699                 ftype = VM_PROT_EXECUTE;
 700         else
 701                 ftype = VM_PROT_READ;
 702
 703         if (map != kernel_map) {
 704                 /*
 705                  * Keep swapout from messing with us during this
 706                  *      critical time.
 707                  */
 708                 PROC_LOCK(p);
 709                 ++p->p_lock;
 710                 PROC_UNLOCK(p);
 711
 712                 /* Fault in the user page: */
 713                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 714
 715                 PROC_LOCK(p);
 716                 --p->p_lock;
 717                 PROC_UNLOCK(p);
 718         } else {
 719                 /*
 720                  * Don't have to worry about process locking or stacks in the
 721                  * kernel.
 722                  */
 723                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 724         }
 725         if (rv == KERN_SUCCESS)
 726                 return (0);
 727 nogo:
 728         if (!usermode) {
 729                 if (td->td_intr_nesting_level == 0 &&
 730                     PCPU_GET(curpcb)->pcb_onfault != NULL) {
 731                         frame->tf_rip = (long)PCPU_GET(curpcb)->pcb_onfault;
 732                         return (0);
 733                 }
 734                 trap_fatal(frame, eva);
 735                 return (-1);
 736         }
 737
 738         return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 739 }
 740
 741 static void
 742 trap_fatal(frame, eva)
 743         struct trapframe *frame;
 744         vm_offset_t eva;
 745 {
 746         int code, ss;
 747         u_int type;
 748         long esp;
 749         struct soft_segment_descriptor softseg;
 750         char *msg;
 751
 752         code = frame->tf_err;
 753         type = frame->tf_trapno;
 754         sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)],
 755             &softseg);
 756
 757         if (type <= MAX_TRAP_MSG)
 758                 msg = trap_msg[type];
 759         else
 760                 msg = "UNKNOWN";
 761         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 762             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
 763 #ifdef SMP
 764         /* two separate prints in case of a trap on an unmapped page */
 765         printf("cpuid = %d; ", PCPU_GET(cpuid));
 766         printf("apic id = %02x\n", PCPU_GET(apic_id));
 767 #endif
 768         if (type == T_PAGEFLT) {
 769                 printf("fault virtual address   = 0x%lx\n", eva);
 770                 printf("fault code              = %s %s %s, %s\n",
 771                         code & PGEX_U ? "user" : "supervisor",
 772                         code & PGEX_W ? "write" : "read",
 773                         code & PGEX_I ? "instruction" : "data",
 774                         code & PGEX_P ? "protection violation" : "page not present");
 775         }
 776         printf("instruction pointer     = 0x%lx:0x%lx\n",
 777                frame->tf_cs & 0xffff, frame->tf_rip);
 778         if (ISPL(frame->tf_cs) == SEL_UPL) {
 779                 ss = frame->tf_ss & 0xffff;
 780                 esp = frame->tf_rsp;
 781         } else {
 782                 ss = GSEL(GDATA_SEL, SEL_KPL);
 783                 esp = (long)&frame->tf_rsp;
 784         }
 785         printf("stack pointer           = 0x%x:0x%lx\n", ss, esp);
 786         printf("frame pointer           = 0x%x:0x%lx\n", ss, frame->tf_rbp);
 787         printf("code segment            = base 0x%lx, limit 0x%lx, type 0x%x\n",
 788                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 789         printf("                        = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
 790                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
 791                softseg.ssd_gran);
 792         printf("processor eflags        = ");
 793         if (frame->tf_rflags & PSL_T)
 794                 printf("trace trap, ");
 795         if (frame->tf_rflags & PSL_I)
 796                 printf("interrupt enabled, ");
 797         if (frame->tf_rflags & PSL_NT)
 798                 printf("nested task, ");
 799         if (frame->tf_rflags & PSL_RF)
 800                 printf("resume, ");
 801         printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
 802         printf("current process         = ");
 803         if (curproc) {
 804                 printf("%lu (%s)\n",
 805                     (u_long)curproc->p_pid, curthread->td_name ?
 806                     curthread->td_name : "");
 807         } else {
 808                 printf("Idle\n");
 809         }
 810
 811 #ifdef KDB
 812         if (debugger_on_panic || kdb_active)
 813                 if (kdb_trap(type, 0, frame))
 814                         return;
 815 #endif
 816         printf("trap number             = %d\n", type);
 817         if (type <= MAX_TRAP_MSG)
 818                 panic("%s", trap_msg[type]);
 819         else
 820                 panic("unknown/reserved trap");
 821 }
 822
 823 /*
 824  * Double fault handler. Called when a fault occurs while writing
 825  * a frame for a trap/exception onto the stack. This usually occurs
 826  * when the stack overflows (such is the case with infinite recursion,
 827  * for example).
 828  */
 829 void
 830 dblfault_handler(struct trapframe *frame)
 831 {
 832 #ifdef KDTRACE_HOOKS
 833         if (dtrace_doubletrap_func != NULL)
 834                 (*dtrace_doubletrap_func)();
 835 #endif
 836         printf("\nFatal double fault\n");
 837         printf("rip = 0x%lx\n", frame->tf_rip);
 838         printf("rsp = 0x%lx\n", frame->tf_rsp);
 839         printf("rbp = 0x%lx\n", frame->tf_rbp);
 840 #ifdef SMP
 841         /* two separate prints in case of a trap on an unmapped page */
 842         printf("cpuid = %d; ", PCPU_GET(cpuid));
 843         printf("apic id = %02x\n", PCPU_GET(apic_id));
 844 #endif
 845         panic("double fault");
 846 }
 847
 848 int
 849 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
 850 {
 851         struct proc *p;
 852         struct trapframe *frame;
 853         register_t *argp;
 854         caddr_t params;
 855         int reg, regcnt, error;
 856
 857         p = td->td_proc;
 858         frame = td->td_frame;
 859         reg = 0;
 860         regcnt = 6;
 861
 862         params = (caddr_t)frame->tf_rsp + sizeof(register_t);
 863         sa->code = frame->tf_rax;
 864
 865         if (sa->code == SYS_syscall || sa->code == SYS___syscall) {
 866                 sa->code = frame->tf_rdi;
 867                 reg++;
 868                 regcnt--;
 869         }
 870         if (p->p_sysent->sv_mask)
 871                 sa->code &= p->p_sysent->sv_mask;
 872
 873         if (sa->code >= p->p_sysent->sv_size)
 874                 sa->callp = &p->p_sysent->sv_table[0];
 875         else
 876                 sa->callp = &p->p_sysent->sv_table[sa->code];
 877
 878         sa->narg = sa->callp->sy_narg;
 879         KASSERT(sa->narg <= sizeof(sa->args) / sizeof(sa->args[0]),
 880             ("Too many syscall arguments!"));
 881         error = 0;
 882         argp = &frame->tf_rdi;
 883         argp += reg;
 884         bcopy(argp, sa->args, sizeof(sa->args[0]) * regcnt);
 885         if (sa->narg > regcnt) {
 886                 KASSERT(params != NULL, ("copyin args with no params!"));
 887                 error = copyin(params, &sa->args[regcnt],
 888                     (sa->narg - regcnt) * sizeof(sa->args[0]));
 889         }
 890
 891         if (error == 0) {
 892                 td->td_retval[0] = 0;
 893                 td->td_retval[1] = frame->tf_rdx;
 894         }
 895
 896         return (error);
 897 }
 898
 899 #include "../../kern/subr_syscall.c"
 900
 901 /*
 902  *      syscall -       system call request C handler
 903  *
 904  *      A system call is essentially treated as a trap.
 905  */
 906 void
 907 amd64_syscall(struct thread *td, int traced)
 908 {
 909         struct syscall_args sa;
 910         int error;
 911         ksiginfo_t ksi;
 912
 913 #ifdef DIAGNOSTIC
 914         if (ISPL(td->td_frame->tf_cs) != SEL_UPL) {
 915                 panic("syscall");
 916                 /* NOT REACHED */
 917         }
 918 #endif
 919         error = syscallenter(td, &sa);
 920
 921         /*
 922          * Traced syscall.
 923          */
 924         if (__predict_false(traced)) {
 925                 td->td_frame->tf_rflags &= ~PSL_T;
 926                 ksiginfo_init_trap(&ksi);
 927                 ksi.ksi_signo = SIGTRAP;
 928                 ksi.ksi_code = TRAP_TRACE;
 929                 ksi.ksi_addr = (void *)td->td_frame->tf_rip;
 930                 trapsignal(td, &ksi);
 931         }
 932
 933         KASSERT(PCB_USER_FPU(td->td_pcb),
 934             ("System call %s returing with kernel FPU ctx leaked",
 935              syscallname(td->td_proc, sa.code)));
 936         KASSERT(td->td_pcb->pcb_save == &td->td_pcb->pcb_user_save,
 937             ("System call %s returning with mangled pcb_save",
 938              syscallname(td->td_proc, sa.code)));
 939
 940         syscallret(td, error, &sa);
 941 }