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            33
 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         "",                                     /* 31 unused (reserved) */
 162         "DTrace pid return trap",               /* 32 T_DTRACE_RET */
 163         "DTrace fasttrap probe trap",           /* 33 T_DTRACE_PROBE */
 164 };
 165
 166 #ifdef KDB
 167 static int kdb_on_nmi = 1;
 168 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
 169         &kdb_on_nmi, 0, "Go to KDB on NMI");
 170 TUNABLE_INT("machdep.kdb_on_nmi", &kdb_on_nmi);
 171 #endif
 172 static int panic_on_nmi = 1;
 173 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
 174         &panic_on_nmi, 0, "Panic on NMI");
 175 TUNABLE_INT("machdep.panic_on_nmi", &panic_on_nmi);
 176 static int prot_fault_translation = 0;
 177 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
 178         &prot_fault_translation, 0, "Select signal to deliver on protection fault");
 179
 180 /*
 181  * Exception, fault, and trap interface to the FreeBSD kernel.
 182  * This common code is called from assembly language IDT gate entry
 183  * routines that prepare a suitable stack frame, and restore this
 184  * frame after the exception has been processed.
 185  */
 186
 187 void
 188 trap(struct trapframe *frame)
 189 {
 190         struct thread *td = curthread;
 191         struct proc *p = td->td_proc;
 192         int i = 0, ucode = 0, code;
 193         u_int type;
 194         register_t addr = 0;
 195         ksiginfo_t ksi;
 196
 197         PCPU_INC(cnt.v_trap);
 198         type = frame->tf_trapno;
 199
 200 #ifdef SMP
 201         /* Handler for NMI IPIs used for stopping CPUs. */
 202         if (type == T_NMI) {
 203                  if (ipi_nmi_handler() == 0)
 204                            goto out;
 205         }
 206 #endif /* SMP */
 207
 208 #ifdef KDB
 209         if (kdb_active) {
 210                 kdb_reenter();
 211                 goto out;
 212         }
 213 #endif
 214
 215         if (type == T_RESERVED) {
 216                 trap_fatal(frame, 0);
 217                 goto out;
 218         }
 219
 220 #ifdef  HWPMC_HOOKS
 221         /*
 222          * CPU PMCs interrupt using an NMI.  If the PMC module is
 223          * active, pass the 'rip' value to the PMC module's interrupt
 224          * handler.  A return value of '1' from the handler means that
 225          * the NMI was handled by it and we can return immediately.
 226          */
 227         if (type == T_NMI && pmc_intr &&
 228             (*pmc_intr)(PCPU_GET(cpuid), frame))
 229                 goto out;
 230 #endif
 231
 232         if (type == T_MCHK) {
 233                 if (!mca_intr())
 234                         trap_fatal(frame, 0);
 235                 goto out;
 236         }
 237
 238 #ifdef KDTRACE_HOOKS
 239         /*
 240          * A trap can occur while DTrace executes a probe. Before
 241          * executing the probe, DTrace blocks re-scheduling and sets
 242          * a flag in it's per-cpu flags to indicate that it doesn't
 243          * want to fault. On returning from the probe, the no-fault
 244          * flag is cleared and finally re-scheduling is enabled.
 245          *
 246          * If the DTrace kernel module has registered a trap handler,
 247          * call it and if it returns non-zero, assume that it has
 248          * handled the trap and modified the trap frame so that this
 249          * function can return normally.
 250          */
 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                 else if (type == T_BPTFLT &&
 261                     dtrace_pid_probe_ptr != NULL &&
 262                     dtrace_pid_probe_ptr(&regs) == 0)
 263                         goto out;
 264                 else if (type == T_DTRACE_RET &&
 265                     dtrace_return_probe_ptr != NULL &&
 266                     dtrace_return_probe_ptr(&regs) == 0)
 267                         goto out;
 268         }
 269         if (dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
 270                 goto out;
 271 #endif
 272
 273         if ((frame->tf_rflags & PSL_I) == 0) {
 274                 /*
 275                  * Buggy application or kernel code has disabled
 276                  * interrupts and then trapped.  Enabling interrupts
 277                  * now is wrong, but it is better than running with
 278                  * interrupts disabled until they are accidentally
 279                  * enabled later.
 280                  */
 281                 if (ISPL(frame->tf_cs) == SEL_UPL)
 282                         uprintf(
 283                             "pid %ld (%s): trap %d with interrupts disabled\n",
 284                             (long)curproc->p_pid, curthread->td_name, type);
 285                 else if (type != T_NMI && type != T_BPTFLT &&
 286                     type != T_TRCTRAP) {
 287                         /*
 288                          * XXX not quite right, since this may be for a
 289                          * multiple fault in user mode.
 290                          */
 291                         printf("kernel trap %d with interrupts disabled\n",
 292                             type);
 293
 294                         /*
 295                          * We shouldn't enable interrupts while holding a
 296                          * spin lock.
 297                          */
 298                         if (td->td_md.md_spinlock_count == 0)
 299                                 enable_intr();
 300                 }
 301         }
 302
 303         code = frame->tf_err;
 304         if (type == T_PAGEFLT) {
 305                 /*
 306                  * If we get a page fault while in a critical section, then
 307                  * it is most likely a fatal kernel page fault.  The kernel
 308                  * is already going to panic trying to get a sleep lock to
 309                  * do the VM lookup, so just consider it a fatal trap so the
 310                  * kernel can print out a useful trap message and even get
 311                  * to the debugger.
 312                  *
 313                  * If we get a page fault while holding a non-sleepable
 314                  * lock, then it is most likely a fatal kernel page fault.
 315                  * If WITNESS is enabled, then it's going to whine about
 316                  * bogus LORs with various VM locks, so just skip to the
 317                  * fatal trap handling directly.
 318                  */
 319                 if (td->td_critnest != 0 ||
 320                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 321                     "Kernel page fault") != 0)
 322                         trap_fatal(frame, frame->tf_addr);
 323         }
 324
 325         if (ISPL(frame->tf_cs) == SEL_UPL) {
 326                 /* user trap */
 327
 328                 td->td_pticks = 0;
 329                 td->td_frame = frame;
 330                 addr = frame->tf_rip;
 331                 if (td->td_ucred != p->p_ucred)
 332                         cred_update_thread(td);
 333
 334                 switch (type) {
 335                 case T_PRIVINFLT:       /* privileged instruction fault */
 336                         i = SIGILL;
 337                         ucode = ILL_PRVOPC;
 338                         break;
 339
 340                 case T_BPTFLT:          /* bpt instruction fault */
 341                 case T_TRCTRAP:         /* trace trap */
 342                         enable_intr();
 343                         frame->tf_rflags &= ~PSL_T;
 344                         i = SIGTRAP;
 345                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 346                         break;
 347
 348                 case T_ARITHTRAP:       /* arithmetic trap */
 349                         ucode = fputrap();
 350                         if (ucode == -1)
 351                                 goto userout;
 352                         i = SIGFPE;
 353                         break;
 354
 355                 case T_PROTFLT:         /* general protection fault */
 356                         i = SIGBUS;
 357                         ucode = BUS_OBJERR;
 358                         break;
 359                 case T_STKFLT:          /* stack fault */
 360                 case T_SEGNPFLT:        /* segment not present fault */
 361                         i = SIGBUS;
 362                         ucode = BUS_ADRERR;
 363                         break;
 364                 case T_TSSFLT:          /* invalid TSS fault */
 365                         i = SIGBUS;
 366                         ucode = BUS_OBJERR;
 367                         break;
 368                 case T_DOUBLEFLT:       /* double fault */
 369                 default:
 370                         i = SIGBUS;
 371                         ucode = BUS_OBJERR;
 372                         break;
 373
 374                 case T_PAGEFLT:         /* page fault */
 375                         addr = frame->tf_addr;
 376                         i = trap_pfault(frame, TRUE);
 377                         if (i == -1)
 378                                 goto userout;
 379                         if (i == 0)
 380                                 goto user;
 381
 382                         if (i == SIGSEGV)
 383                                 ucode = SEGV_MAPERR;
 384                         else {
 385                                 if (prot_fault_translation == 0) {
 386                                         /*
 387                                          * Autodetect.
 388                                          * This check also covers the images
 389                                          * without the ABI-tag ELF note.
 390                                          */
 391                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 392                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 393                                                 i = SIGSEGV;
 394                                                 ucode = SEGV_ACCERR;
 395                                         } else {
 396                                                 i = SIGBUS;
 397                                                 ucode = BUS_PAGE_FAULT;
 398                                         }
 399                                 } else if (prot_fault_translation == 1) {
 400                                         /*
 401                                          * Always compat mode.
 402                                          */
 403                                         i = SIGBUS;
 404                                         ucode = BUS_PAGE_FAULT;
 405                                 } else {
 406                                         /*
 407                                          * Always SIGSEGV mode.
 408                                          */
 409                                         i = SIGSEGV;
 410                                         ucode = SEGV_ACCERR;
 411                                 }
 412                         }
 413                         break;
 414
 415                 case T_DIVIDE:          /* integer divide fault */
 416                         ucode = FPE_INTDIV;
 417                         i = SIGFPE;
 418                         break;
 419
 420 #ifdef DEV_ISA
 421                 case T_NMI:
 422                         /* machine/parity/power fail/"kitchen sink" faults */
 423                         if (isa_nmi(code) == 0) {
 424 #ifdef KDB
 425                                 /*
 426                                  * NMI can be hooked up to a pushbutton
 427                                  * for debugging.
 428                                  */
 429                                 if (kdb_on_nmi) {
 430                                         printf ("NMI ... going to debugger\n");
 431                                         kdb_trap(type, 0, frame);
 432                                 }
 433 #endif /* KDB */
 434                                 goto userout;
 435                         } else if (panic_on_nmi)
 436                                 panic("NMI indicates hardware failure");
 437                         break;
 438 #endif /* DEV_ISA */
 439
 440                 case T_OFLOW:           /* integer overflow fault */
 441                         ucode = FPE_INTOVF;
 442                         i = SIGFPE;
 443                         break;
 444
 445                 case T_BOUND:           /* bounds check fault */
 446                         ucode = FPE_FLTSUB;
 447                         i = SIGFPE;
 448                         break;
 449
 450                 case T_DNA:
 451                         /* transparent fault (due to context switch "late") */
 452                         KASSERT(PCB_USER_FPU(td->td_pcb),
 453                             ("kernel FPU ctx has leaked"));
 454                         fpudna();
 455                         goto userout;
 456
 457                 case T_FPOPFLT:         /* FPU operand fetch fault */
 458                         ucode = ILL_COPROC;
 459                         i = SIGILL;
 460                         break;
 461
 462                 case T_XMMFLT:          /* SIMD floating-point exception */
 463                         ucode = 0; /* XXX */
 464                         i = SIGFPE;
 465                         break;
 466                 }
 467         } else {
 468                 /* kernel trap */
 469
 470                 KASSERT(cold || td->td_ucred != NULL,
 471                     ("kernel trap doesn't have ucred"));
 472                 switch (type) {
 473                 case T_PAGEFLT:                 /* page fault */
 474                         (void) trap_pfault(frame, FALSE);
 475                         goto out;
 476
 477                 case T_DNA:
 478                         KASSERT(!PCB_USER_FPU(td->td_pcb),
 479                             ("Unregistered use of FPU in kernel"));
 480                         fpudna();
 481                         goto out;
 482
 483                 case T_ARITHTRAP:       /* arithmetic trap */
 484                 case T_XMMFLT:          /* SIMD floating-point exception */
 485                 case T_FPOPFLT:         /* FPU operand fetch fault */
 486                         /*
 487                          * XXXKIB for now disable any FPU traps in kernel
 488                          * handler registration seems to be overkill
 489                          */
 490                         trap_fatal(frame, 0);
 491                         goto out;
 492
 493                 case T_STKFLT:          /* stack fault */
 494                         break;
 495
 496                 case T_PROTFLT:         /* general protection fault */
 497                 case T_SEGNPFLT:        /* segment not present fault */
 498                         if (td->td_intr_nesting_level != 0)
 499                                 break;
 500
 501                         /*
 502                          * Invalid segment selectors and out of bounds
 503                          * %rip's and %rsp's can be set up in user mode.
 504                          * This causes a fault in kernel mode when the
 505                          * kernel tries to return to user mode.  We want
 506                          * to get this fault so that we can fix the
 507                          * problem here and not have to check all the
 508                          * selectors and pointers when the user changes
 509                          * them.
 510                          */
 511                         if (frame->tf_rip == (long)doreti_iret) {
 512                                 frame->tf_rip = (long)doreti_iret_fault;
 513                                 goto out;
 514                         }
 515                         if (frame->tf_rip == (long)ld_ds) {
 516                                 frame->tf_rip = (long)ds_load_fault;
 517                                 goto out;
 518                         }
 519                         if (frame->tf_rip == (long)ld_es) {
 520                                 frame->tf_rip = (long)es_load_fault;
 521                                 goto out;
 522                         }
 523                         if (frame->tf_rip == (long)ld_fs) {
 524                                 frame->tf_rip = (long)fs_load_fault;
 525                                 goto out;
 526                         }
 527                         if (frame->tf_rip == (long)ld_gs) {
 528                                 frame->tf_rip = (long)gs_load_fault;
 529                                 goto out;
 530                         }
 531                         if (frame->tf_rip == (long)ld_gsbase) {
 532                                 frame->tf_rip = (long)gsbase_load_fault;
 533                                 goto out;
 534                         }
 535                         if (frame->tf_rip == (long)ld_fsbase) {
 536                                 frame->tf_rip = (long)fsbase_load_fault;
 537                                 goto out;
 538                         }
 539                         if (PCPU_GET(curpcb)->pcb_onfault != NULL) {
 540                                 frame->tf_rip =
 541                                     (long)PCPU_GET(curpcb)->pcb_onfault;
 542                                 goto out;
 543                         }
 544                         break;
 545
 546                 case T_TSSFLT:
 547                         /*
 548                          * PSL_NT can be set in user mode and isn't cleared
 549                          * automatically when the kernel is entered.  This
 550                          * causes a TSS fault when the kernel attempts to
 551                          * `iret' because the TSS link is uninitialized.  We
 552                          * want to get this fault so that we can fix the
 553                          * problem here and not every time the kernel is
 554                          * entered.
 555                          */
 556                         if (frame->tf_rflags & PSL_NT) {
 557                                 frame->tf_rflags &= ~PSL_NT;
 558                                 goto out;
 559                         }
 560                         break;
 561
 562                 case T_TRCTRAP:  /* trace trap */
 563                         /*
 564                          * Ignore debug register trace traps due to
 565                          * accesses in the user's address space, which
 566                          * can happen under several conditions such as
 567                          * if a user sets a watchpoint on a buffer and
 568                          * then passes that buffer to a system call.
 569                          * We still want to get TRCTRAPS for addresses
 570                          * in kernel space because that is useful when
 571                          * debugging the kernel.
 572                          */
 573                         if (user_dbreg_trap()) {
 574                                 /*
 575                                  * Reset breakpoint bits because the
 576                                  * processor doesn't
 577                                  */
 578                                 /* XXX check upper bits here */
 579                                 load_dr6(rdr6() & 0xfffffff0);
 580                                 goto out;
 581                         }
 582                         /*
 583                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 584                          */
 585                 case T_BPTFLT:
 586                         /*
 587                          * If KDB is enabled, let it handle the debugger trap.
 588                          * Otherwise, debugger traps "can't happen".
 589                          */
 590 #ifdef KDB
 591                         if (kdb_trap(type, 0, frame))
 592                                 goto out;
 593 #endif
 594                         break;
 595
 596 #ifdef DEV_ISA
 597                 case T_NMI:
 598                         /* machine/parity/power fail/"kitchen sink" faults */
 599                         if (isa_nmi(code) == 0) {
 600 #ifdef KDB
 601                                 /*
 602                                  * NMI can be hooked up to a pushbutton
 603                                  * for debugging.
 604                                  */
 605                                 if (kdb_on_nmi) {
 606                                         printf ("NMI ... going to debugger\n");
 607                                         kdb_trap(type, 0, frame);
 608                                 }
 609 #endif /* KDB */
 610                                 goto out;
 611                         } else if (panic_on_nmi == 0)
 612                                 goto out;
 613                         /* FALLTHROUGH */
 614 #endif /* DEV_ISA */
 615                 }
 616
 617                 trap_fatal(frame, 0);
 618                 goto out;
 619         }
 620
 621         /* Translate fault for emulators (e.g. Linux) */
 622         if (*p->p_sysent->sv_transtrap)
 623                 i = (*p->p_sysent->sv_transtrap)(i, type);
 624
 625         ksiginfo_init_trap(&ksi);
 626         ksi.ksi_signo = i;
 627         ksi.ksi_code = ucode;
 628         ksi.ksi_trapno = type;
 629         ksi.ksi_addr = (void *)addr;
 630         trapsignal(td, &ksi);
 631
 632 user:
 633         userret(td, frame);
 634         mtx_assert(&Giant, MA_NOTOWNED);
 635         KASSERT(PCB_USER_FPU(td->td_pcb),
 636             ("Return from trap with kernel FPU ctx leaked"));
 637 userout:
 638 out:
 639         return;
 640 }
 641
 642 static int
 643 trap_pfault(frame, usermode)
 644         struct trapframe *frame;
 645         int usermode;
 646 {
 647         vm_offset_t va;
 648         struct vmspace *vm = NULL;
 649         vm_map_t map;
 650         int rv = 0;
 651         vm_prot_t ftype;
 652         struct thread *td = curthread;
 653         struct proc *p = td->td_proc;
 654         vm_offset_t eva = frame->tf_addr;
 655
 656         va = trunc_page(eva);
 657         if (va >= VM_MIN_KERNEL_ADDRESS) {
 658                 /*
 659                  * Don't allow user-mode faults in kernel address space.
 660                  */
 661                 if (usermode)
 662                         goto nogo;
 663
 664                 map = kernel_map;
 665         } else {
 666                 /*
 667                  * This is a fault on non-kernel virtual memory.
 668                  * vm is initialized above to NULL. If curproc is NULL
 669                  * or curproc->p_vmspace is NULL the fault is fatal.
 670                  */
 671                 if (p != NULL)
 672                         vm = p->p_vmspace;
 673
 674                 if (vm == NULL)
 675                         goto nogo;
 676
 677                 map = &vm->vm_map;
 678
 679                 /*
 680                  * When accessing a usermode address, kernel must be
 681                  * ready to accept the page fault, and provide a
 682                  * handling routine.  Since accessing the address
 683                  * without the handler is a bug, do not try to handle
 684                  * it normally, and panic immediately.
 685                  */
 686                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 687                     PCPU_GET(curpcb)->pcb_onfault == NULL)) {
 688                         trap_fatal(frame, eva);
 689                         return (-1);
 690                 }
 691         }
 692
 693         /*
 694          * PGEX_I is defined only if the execute disable bit capability is
 695          * supported and enabled.
 696          */
 697         if (frame->tf_err & PGEX_W)
 698                 ftype = VM_PROT_WRITE;
 699         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 700                 ftype = VM_PROT_EXECUTE;
 701         else
 702                 ftype = VM_PROT_READ;
 703
 704         if (map != kernel_map) {
 705                 /*
 706                  * Keep swapout from messing with us during this
 707                  *      critical time.
 708                  */
 709                 PROC_LOCK(p);
 710                 ++p->p_lock;
 711                 PROC_UNLOCK(p);
 712
 713                 /* Fault in the user page: */
 714                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 715
 716                 PROC_LOCK(p);
 717                 --p->p_lock;
 718                 PROC_UNLOCK(p);
 719         } else {
 720                 /*
 721                  * Don't have to worry about process locking or stacks in the
 722                  * kernel.
 723                  */
 724                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 725         }
 726         if (rv == KERN_SUCCESS)
 727                 return (0);
 728 nogo:
 729         if (!usermode) {
 730                 if (td->td_intr_nesting_level == 0 &&
 731                     PCPU_GET(curpcb)->pcb_onfault != NULL) {
 732                         frame->tf_rip = (long)PCPU_GET(curpcb)->pcb_onfault;
 733                         return (0);
 734                 }
 735                 trap_fatal(frame, eva);
 736                 return (-1);
 737         }
 738
 739         return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 740 }
 741
 742 static void
 743 trap_fatal(frame, eva)
 744         struct trapframe *frame;
 745         vm_offset_t eva;
 746 {
 747         int code, ss;
 748         u_int type;
 749         long esp;
 750         struct soft_segment_descriptor softseg;
 751         char *msg;
 752
 753         code = frame->tf_err;
 754         type = frame->tf_trapno;
 755         sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)],
 756             &softseg);
 757
 758         if (type <= MAX_TRAP_MSG)
 759                 msg = trap_msg[type];
 760         else
 761                 msg = "UNKNOWN";
 762         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 763             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
 764 #ifdef SMP
 765         /* two separate prints in case of a trap on an unmapped page */
 766         printf("cpuid = %d; ", PCPU_GET(cpuid));
 767         printf("apic id = %02x\n", PCPU_GET(apic_id));
 768 #endif
 769         if (type == T_PAGEFLT) {
 770                 printf("fault virtual address   = 0x%lx\n", eva);
 771                 printf("fault code              = %s %s %s, %s\n",
 772                         code & PGEX_U ? "user" : "supervisor",
 773                         code & PGEX_W ? "write" : "read",
 774                         code & PGEX_I ? "instruction" : "data",
 775                         code & PGEX_P ? "protection violation" : "page not present");
 776         }
 777         printf("instruction pointer     = 0x%lx:0x%lx\n",
 778                frame->tf_cs & 0xffff, frame->tf_rip);
 779         if (ISPL(frame->tf_cs) == SEL_UPL) {
 780                 ss = frame->tf_ss & 0xffff;
 781                 esp = frame->tf_rsp;
 782         } else {
 783                 ss = GSEL(GDATA_SEL, SEL_KPL);
 784                 esp = (long)&frame->tf_rsp;
 785         }
 786         printf("stack pointer           = 0x%x:0x%lx\n", ss, esp);
 787         printf("frame pointer           = 0x%x:0x%lx\n", ss, frame->tf_rbp);
 788         printf("code segment            = base 0x%lx, limit 0x%lx, type 0x%x\n",
 789                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 790         printf("                        = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
 791                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
 792                softseg.ssd_gran);
 793         printf("processor eflags        = ");
 794         if (frame->tf_rflags & PSL_T)
 795                 printf("trace trap, ");
 796         if (frame->tf_rflags & PSL_I)
 797                 printf("interrupt enabled, ");
 798         if (frame->tf_rflags & PSL_NT)
 799                 printf("nested task, ");
 800         if (frame->tf_rflags & PSL_RF)
 801                 printf("resume, ");
 802         printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
 803         printf("current process         = ");
 804         if (curproc) {
 805                 printf("%lu (%s)\n",
 806                     (u_long)curproc->p_pid, curthread->td_name ?
 807                     curthread->td_name : "");
 808         } else {
 809                 printf("Idle\n");
 810         }
 811
 812 #ifdef KDB
 813         if (debugger_on_panic || kdb_active)
 814                 if (kdb_trap(type, 0, frame))
 815                         return;
 816 #endif
 817         printf("trap number             = %d\n", type);
 818         if (type <= MAX_TRAP_MSG)
 819                 panic("%s", trap_msg[type]);
 820         else
 821                 panic("unknown/reserved trap");
 822 }
 823
 824 /*
 825  * Double fault handler. Called when a fault occurs while writing
 826  * a frame for a trap/exception onto the stack. This usually occurs
 827  * when the stack overflows (such is the case with infinite recursion,
 828  * for example).
 829  */
 830 void
 831 dblfault_handler(struct trapframe *frame)
 832 {
 833 #ifdef KDTRACE_HOOKS
 834         if (dtrace_doubletrap_func != NULL)
 835                 (*dtrace_doubletrap_func)();
 836 #endif
 837         printf("\nFatal double fault\n");
 838         printf("rip = 0x%lx\n", frame->tf_rip);
 839         printf("rsp = 0x%lx\n", frame->tf_rsp);
 840         printf("rbp = 0x%lx\n", frame->tf_rbp);
 841 #ifdef SMP
 842         /* two separate prints in case of a trap on an unmapped page */
 843         printf("cpuid = %d; ", PCPU_GET(cpuid));
 844         printf("apic id = %02x\n", PCPU_GET(apic_id));
 845 #endif
 846         panic("double fault");
 847 }
 848
 849 int
 850 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
 851 {
 852         struct proc *p;
 853         struct trapframe *frame;
 854         register_t *argp;
 855         caddr_t params;
 856         int reg, regcnt, error;
 857
 858         p = td->td_proc;
 859         frame = td->td_frame;
 860         reg = 0;
 861         regcnt = 6;
 862
 863         params = (caddr_t)frame->tf_rsp + sizeof(register_t);
 864         sa->code = frame->tf_rax;
 865
 866         if (sa->code == SYS_syscall || sa->code == SYS___syscall) {
 867                 sa->code = frame->tf_rdi;
 868                 reg++;
 869                 regcnt--;
 870         }
 871         if (p->p_sysent->sv_mask)
 872                 sa->code &= p->p_sysent->sv_mask;
 873
 874         if (sa->code >= p->p_sysent->sv_size)
 875                 sa->callp = &p->p_sysent->sv_table[0];
 876         else
 877                 sa->callp = &p->p_sysent->sv_table[sa->code];
 878
 879         sa->narg = sa->callp->sy_narg;
 880         KASSERT(sa->narg <= sizeof(sa->args) / sizeof(sa->args[0]),
 881             ("Too many syscall arguments!"));
 882         error = 0;
 883         argp = &frame->tf_rdi;
 884         argp += reg;
 885         bcopy(argp, sa->args, sizeof(sa->args[0]) * regcnt);
 886         if (sa->narg > regcnt) {
 887                 KASSERT(params != NULL, ("copyin args with no params!"));
 888                 error = copyin(params, &sa->args[regcnt],
 889                     (sa->narg - regcnt) * sizeof(sa->args[0]));
 890         }
 891
 892         if (error == 0) {
 893                 td->td_retval[0] = 0;
 894                 td->td_retval[1] = frame->tf_rdx;
 895         }
 896
 897         return (error);
 898 }
 899
 900 #include "../../kern/subr_syscall.c"
 901
 902 /*
 903  *      syscall -       system call request C handler
 904  *
 905  *      A system call is essentially treated as a trap.
 906  */
 907 void
 908 amd64_syscall(struct thread *td, int traced)
 909 {
 910         struct syscall_args sa;
 911         int error;
 912         ksiginfo_t ksi;
 913
 914 #ifdef DIAGNOSTIC
 915         if (ISPL(td->td_frame->tf_cs) != SEL_UPL) {
 916                 panic("syscall");
 917                 /* NOT REACHED */
 918         }
 919 #endif
 920         error = syscallenter(td, &sa);
 921
 922         /*
 923          * Traced syscall.
 924          */
 925         if (__predict_false(traced)) {
 926                 td->td_frame->tf_rflags &= ~PSL_T;
 927                 ksiginfo_init_trap(&ksi);
 928                 ksi.ksi_signo = SIGTRAP;
 929                 ksi.ksi_code = TRAP_TRACE;
 930                 ksi.ksi_addr = (void *)td->td_frame->tf_rip;
 931                 trapsignal(td, &ksi);
 932         }
 933
 934         KASSERT(PCB_USER_FPU(td->td_pcb),
 935             ("System call %s returing with kernel FPU ctx leaked",
 936              syscallname(td->td_proc, sa.code)));
 937         KASSERT(td->td_pcb->pcb_save == &td->td_pcb->pcb_user_save,
 938             ("System call %s returning with mangled pcb_save",
 939              syscallname(td->td_proc, sa.code)));
 940
 941         syscallret(td, error, &sa);
 942
 943         /*
 944          * If the user-supplied value of %rip is not a canonical
 945          * address, then some CPUs will trigger a ring 0 #GP during
 946          * the sysret instruction.  However, the fault handler would
 947          * execute with the user's %gs and %rsp in ring 0 which would
 948          * not be safe.  Instead, preemptively kill the thread with a
 949          * SIGBUS.
 950          */
 951         if (td->td_frame->tf_rip >= VM_MAXUSER_ADDRESS) {
 952                 ksiginfo_init_trap(&ksi);
 953                 ksi.ksi_signo = SIGBUS;
 954                 ksi.ksi_code = BUS_OBJERR;
 955                 ksi.ksi_trapno = T_PROTFLT;
 956                 ksi.ksi_addr = (void *)td->td_frame->tf_rip;
 957                 trapsignal(td, &ksi);
 958         }
 959 }