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