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 PMC_SOFT_DEFINE( , , page_fault, all);
  75 PMC_SOFT_DEFINE( , , page_fault, read);
  76 PMC_SOFT_DEFINE( , , page_fault, write);
  77 #endif
  78
  79 #include <vm/vm.h>
  80 #include <vm/vm_param.h>
  81 #include <vm/pmap.h>
  82 #include <vm/vm_kern.h>
  83 #include <vm/vm_map.h>
  84 #include <vm/vm_page.h>
  85 #include <vm/vm_extern.h>
  86
  87 #include <machine/cpu.h>
  88 #include <machine/intr_machdep.h>
  89 #include <x86/mca.h>
  90 #include <machine/md_var.h>
  91 #include <machine/pcb.h>
  92 #ifdef SMP
  93 #include <machine/smp.h>
  94 #endif
  95 #include <machine/tss.h>
  96
  97 #ifdef KDTRACE_HOOKS
  98 #include <sys/dtrace_bsd.h>
  99 #endif
 100
 101 extern void trap(struct trapframe *frame);
 102 extern void syscall(struct trapframe *frame);
 103 void dblfault_handler(struct trapframe *frame);
 104
 105 static int trap_pfault(struct trapframe *, int);
 106 static void trap_fatal(struct trapframe *, vm_offset_t);
 107
 108 #define MAX_TRAP_MSG            32
 109 static char *trap_msg[] = {
 110         "",                                     /*  0 unused */
 111         "privileged instruction fault",         /*  1 T_PRIVINFLT */
 112         "",                                     /*  2 unused */
 113         "breakpoint instruction fault",         /*  3 T_BPTFLT */
 114         "",                                     /*  4 unused */
 115         "",                                     /*  5 unused */
 116         "arithmetic trap",                      /*  6 T_ARITHTRAP */
 117         "",                                     /*  7 unused */
 118         "",                                     /*  8 unused */
 119         "general protection fault",             /*  9 T_PROTFLT */
 120         "trace trap",                           /* 10 T_TRCTRAP */
 121         "",                                     /* 11 unused */
 122         "page fault",                           /* 12 T_PAGEFLT */
 123         "",                                     /* 13 unused */
 124         "alignment fault",                      /* 14 T_ALIGNFLT */
 125         "",                                     /* 15 unused */
 126         "",                                     /* 16 unused */
 127         "",                                     /* 17 unused */
 128         "integer divide fault",                 /* 18 T_DIVIDE */
 129         "non-maskable interrupt trap",          /* 19 T_NMI */
 130         "overflow trap",                        /* 20 T_OFLOW */
 131         "FPU bounds check fault",               /* 21 T_BOUND */
 132         "FPU device not available",             /* 22 T_DNA */
 133         "double fault",                         /* 23 T_DOUBLEFLT */
 134         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
 135         "invalid TSS fault",                    /* 25 T_TSSFLT */
 136         "segment not present fault",            /* 26 T_SEGNPFLT */
 137         "stack fault",                          /* 27 T_STKFLT */
 138         "machine check trap",                   /* 28 T_MCHK */
 139         "SIMD floating-point exception",        /* 29 T_XMMFLT */
 140         "reserved (unknown) fault",             /* 30 T_RESERVED */
 141         "",                                     /* 31 unused (reserved) */
 142         "DTrace pid return trap",               /* 32 T_DTRACE_RET */
 143 };
 144
 145 #ifdef KDB
 146 static int kdb_on_nmi = 1;
 147 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
 148         &kdb_on_nmi, 0, "Go to KDB on NMI");
 149 TUNABLE_INT("machdep.kdb_on_nmi", &kdb_on_nmi);
 150 #endif
 151 static int panic_on_nmi = 1;
 152 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
 153         &panic_on_nmi, 0, "Panic on NMI");
 154 TUNABLE_INT("machdep.panic_on_nmi", &panic_on_nmi);
 155 static int prot_fault_translation;
 156 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
 157     &prot_fault_translation, 0,
 158     "Select signal to deliver on protection fault");
 159 static int uprintf_signal;
 160 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW,
 161     &uprintf_signal, 0,
 162     "Print debugging information on trap signal to ctty");
 163
 164 /*
 165  * Exception, fault, and trap interface to the FreeBSD kernel.
 166  * This common code is called from assembly language IDT gate entry
 167  * routines that prepare a suitable stack frame, and restore this
 168  * frame after the exception has been processed.
 169  */
 170
 171 void
 172 trap(struct trapframe *frame)
 173 {
 174 #ifdef KDTRACE_HOOKS
 175         struct reg regs;
 176 #endif
 177         struct thread *td = curthread;
 178         struct proc *p = td->td_proc;
 179         int i = 0, ucode = 0, code;
 180         u_int type;
 181         register_t addr = 0;
 182         ksiginfo_t ksi;
 183
 184         PCPU_INC(cnt.v_trap);
 185         type = frame->tf_trapno;
 186
 187 #ifdef SMP
 188         /* Handler for NMI IPIs used for stopping CPUs. */
 189         if (type == T_NMI) {
 190                  if (ipi_nmi_handler() == 0)
 191                            goto out;
 192         }
 193 #endif /* SMP */
 194
 195 #ifdef KDB
 196         if (kdb_active) {
 197                 kdb_reenter();
 198                 goto out;
 199         }
 200 #endif
 201
 202         if (type == T_RESERVED) {
 203                 trap_fatal(frame, 0);
 204                 goto out;
 205         }
 206
 207 #ifdef  HWPMC_HOOKS
 208         /*
 209          * CPU PMCs interrupt using an NMI.  If the PMC module is
 210          * active, pass the 'rip' value to the PMC module's interrupt
 211          * handler.  A return value of '1' from the handler means that
 212          * the NMI was handled by it and we can return immediately.
 213          */
 214         if (type == T_NMI && pmc_intr &&
 215             (*pmc_intr)(PCPU_GET(cpuid), frame))
 216                 goto out;
 217 #endif
 218
 219         if (type == T_MCHK) {
 220                 mca_intr();
 221                 goto out;
 222         }
 223
 224 #ifdef KDTRACE_HOOKS
 225         /*
 226          * A trap can occur while DTrace executes a probe. Before
 227          * executing the probe, DTrace blocks re-scheduling and sets
 228          * a flag in its per-cpu flags to indicate that it doesn't
 229          * want to fault. On returning from the probe, the no-fault
 230          * flag is cleared and finally re-scheduling is enabled.
 231          */
 232         if (dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
 233                 goto out;
 234 #endif
 235
 236         if ((frame->tf_rflags & PSL_I) == 0) {
 237                 /*
 238                  * Buggy application or kernel code has disabled
 239                  * interrupts and then trapped.  Enabling interrupts
 240                  * now is wrong, but it is better than running with
 241                  * interrupts disabled until they are accidentally
 242                  * enabled later.
 243                  */
 244                 if (ISPL(frame->tf_cs) == SEL_UPL)
 245                         uprintf(
 246                             "pid %ld (%s): trap %d with interrupts disabled\n",
 247                             (long)curproc->p_pid, curthread->td_name, type);
 248                 else if (type != T_NMI && type != T_BPTFLT &&
 249                     type != T_TRCTRAP) {
 250                         /*
 251                          * XXX not quite right, since this may be for a
 252                          * multiple fault in user mode.
 253                          */
 254                         printf("kernel trap %d with interrupts disabled\n",
 255                             type);
 256
 257                         /*
 258                          * We shouldn't enable interrupts while holding a
 259                          * spin lock.
 260                          */
 261                         if (td->td_md.md_spinlock_count == 0)
 262                                 enable_intr();
 263                 }
 264         }
 265
 266         code = frame->tf_err;
 267
 268         if (ISPL(frame->tf_cs) == SEL_UPL) {
 269                 /* user trap */
 270
 271                 td->td_pticks = 0;
 272                 td->td_frame = frame;
 273                 addr = frame->tf_rip;
 274                 if (td->td_ucred != p->p_ucred)
 275                         cred_update_thread(td);
 276
 277                 switch (type) {
 278                 case T_PRIVINFLT:       /* privileged instruction fault */
 279                         i = SIGILL;
 280                         ucode = ILL_PRVOPC;
 281                         break;
 282
 283                 case T_BPTFLT:          /* bpt instruction fault */
 284                 case T_TRCTRAP:         /* trace trap */
 285                         enable_intr();
 286 #ifdef KDTRACE_HOOKS
 287                         if (type == T_BPTFLT) {
 288                                 fill_frame_regs(frame, &regs);
 289                                 if (dtrace_pid_probe_ptr != NULL &&
 290                                     dtrace_pid_probe_ptr(&regs) == 0)
 291                                         goto out;
 292                         }
 293 #endif
 294                         frame->tf_rflags &= ~PSL_T;
 295                         i = SIGTRAP;
 296                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 297                         break;
 298
 299                 case T_ARITHTRAP:       /* arithmetic trap */
 300                         ucode = fputrap_x87();
 301                         if (ucode == -1)
 302                                 goto userout;
 303                         i = SIGFPE;
 304                         break;
 305
 306                 case T_PROTFLT:         /* general protection fault */
 307                         i = SIGBUS;
 308                         ucode = BUS_OBJERR;
 309                         break;
 310                 case T_STKFLT:          /* stack fault */
 311                 case T_SEGNPFLT:        /* segment not present fault */
 312                         i = SIGBUS;
 313                         ucode = BUS_ADRERR;
 314                         break;
 315                 case T_TSSFLT:          /* invalid TSS fault */
 316                         i = SIGBUS;
 317                         ucode = BUS_OBJERR;
 318                         break;
 319                 case T_ALIGNFLT:
 320                         i = SIGBUS;
 321                         ucode = BUS_ADRALN;
 322                         break;
 323                 case T_DOUBLEFLT:       /* double fault */
 324                 default:
 325                         i = SIGBUS;
 326                         ucode = BUS_OBJERR;
 327                         break;
 328
 329                 case T_PAGEFLT:         /* page fault */
 330                         addr = frame->tf_addr;
 331                         i = trap_pfault(frame, TRUE);
 332                         if (i == -1)
 333                                 goto userout;
 334                         if (i == 0)
 335                                 goto user;
 336
 337                         if (i == SIGSEGV)
 338                                 ucode = SEGV_MAPERR;
 339                         else {
 340                                 if (prot_fault_translation == 0) {
 341                                         /*
 342                                          * Autodetect.
 343                                          * This check also covers the images
 344                                          * without the ABI-tag ELF note.
 345                                          */
 346                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 347                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 348                                                 i = SIGSEGV;
 349                                                 ucode = SEGV_ACCERR;
 350                                         } else {
 351                                                 i = SIGBUS;
 352                                                 ucode = BUS_PAGE_FAULT;
 353                                         }
 354                                 } else if (prot_fault_translation == 1) {
 355                                         /*
 356                                          * Always compat mode.
 357                                          */
 358                                         i = SIGBUS;
 359                                         ucode = BUS_PAGE_FAULT;
 360                                 } else {
 361                                         /*
 362                                          * Always SIGSEGV mode.
 363                                          */
 364                                         i = SIGSEGV;
 365                                         ucode = SEGV_ACCERR;
 366                                 }
 367                         }
 368                         break;
 369
 370                 case T_DIVIDE:          /* integer divide fault */
 371                         ucode = FPE_INTDIV;
 372                         i = SIGFPE;
 373                         break;
 374
 375 #ifdef DEV_ISA
 376                 case T_NMI:
 377                         /* machine/parity/power fail/"kitchen sink" faults */
 378                         if (isa_nmi(code) == 0) {
 379 #ifdef KDB
 380                                 /*
 381                                  * NMI can be hooked up to a pushbutton
 382                                  * for debugging.
 383                                  */
 384                                 if (kdb_on_nmi) {
 385                                         printf ("NMI ... going to debugger\n");
 386                                         kdb_trap(type, 0, frame);
 387                                 }
 388 #endif /* KDB */
 389                                 goto userout;
 390                         } else if (panic_on_nmi)
 391                                 panic("NMI indicates hardware failure");
 392                         break;
 393 #endif /* DEV_ISA */
 394
 395                 case T_OFLOW:           /* integer overflow fault */
 396                         ucode = FPE_INTOVF;
 397                         i = SIGFPE;
 398                         break;
 399
 400                 case T_BOUND:           /* bounds check fault */
 401                         ucode = FPE_FLTSUB;
 402                         i = SIGFPE;
 403                         break;
 404
 405                 case T_DNA:
 406                         /* transparent fault (due to context switch "late") */
 407                         KASSERT(PCB_USER_FPU(td->td_pcb),
 408                             ("kernel FPU ctx has leaked"));
 409                         fpudna();
 410                         goto userout;
 411
 412                 case T_FPOPFLT:         /* FPU operand fetch fault */
 413                         ucode = ILL_COPROC;
 414                         i = SIGILL;
 415                         break;
 416
 417                 case T_XMMFLT:          /* SIMD floating-point exception */
 418                         ucode = fputrap_sse();
 419                         if (ucode == -1)
 420                                 goto userout;
 421                         i = SIGFPE;
 422                         break;
 423 #ifdef KDTRACE_HOOKS
 424                 case T_DTRACE_RET:
 425                         enable_intr();
 426                         fill_frame_regs(frame, &regs);
 427                         if (dtrace_return_probe_ptr != NULL &&
 428                             dtrace_return_probe_ptr(&regs) == 0)
 429                                 goto out;
 430                         break;
 431 #endif
 432                 }
 433         } else {
 434                 /* kernel trap */
 435
 436                 KASSERT(cold || td->td_ucred != NULL,
 437                     ("kernel trap doesn't have ucred"));
 438                 switch (type) {
 439                 case T_PAGEFLT:                 /* page fault */
 440                         (void) trap_pfault(frame, FALSE);
 441                         goto out;
 442
 443                 case T_DNA:
 444                         KASSERT(!PCB_USER_FPU(td->td_pcb),
 445                             ("Unregistered use of FPU in kernel"));
 446                         fpudna();
 447                         goto out;
 448
 449                 case T_ARITHTRAP:       /* arithmetic trap */
 450                 case T_XMMFLT:          /* SIMD floating-point exception */
 451                 case T_FPOPFLT:         /* FPU operand fetch fault */
 452                         /*
 453                          * XXXKIB for now disable any FPU traps in kernel
 454                          * handler registration seems to be overkill
 455                          */
 456                         trap_fatal(frame, 0);
 457                         goto out;
 458
 459                 case T_STKFLT:          /* stack fault */
 460                         break;
 461
 462                 case T_PROTFLT:         /* general protection fault */
 463                 case T_SEGNPFLT:        /* segment not present fault */
 464                         if (td->td_intr_nesting_level != 0)
 465                                 break;
 466
 467                         /*
 468                          * Invalid segment selectors and out of bounds
 469                          * %rip's and %rsp's can be set up in user mode.
 470                          * This causes a fault in kernel mode when the
 471                          * kernel tries to return to user mode.  We want
 472                          * to get this fault so that we can fix the
 473                          * problem here and not have to check all the
 474                          * selectors and pointers when the user changes
 475                          * them.
 476                          */
 477                         if (frame->tf_rip == (long)doreti_iret) {
 478                                 frame->tf_rip = (long)doreti_iret_fault;
 479                                 goto out;
 480                         }
 481                         if (frame->tf_rip == (long)ld_ds) {
 482                                 frame->tf_rip = (long)ds_load_fault;
 483                                 goto out;
 484                         }
 485                         if (frame->tf_rip == (long)ld_es) {
 486                                 frame->tf_rip = (long)es_load_fault;
 487                                 goto out;
 488                         }
 489                         if (frame->tf_rip == (long)ld_fs) {
 490                                 frame->tf_rip = (long)fs_load_fault;
 491                                 goto out;
 492                         }
 493                         if (frame->tf_rip == (long)ld_gs) {
 494                                 frame->tf_rip = (long)gs_load_fault;
 495                                 goto out;
 496                         }
 497                         if (frame->tf_rip == (long)ld_gsbase) {
 498                                 frame->tf_rip = (long)gsbase_load_fault;
 499                                 goto out;
 500                         }
 501                         if (frame->tf_rip == (long)ld_fsbase) {
 502                                 frame->tf_rip = (long)fsbase_load_fault;
 503                                 goto out;
 504                         }
 505                         if (curpcb->pcb_onfault != NULL) {
 506                                 frame->tf_rip = (long)curpcb->pcb_onfault;
 507                                 goto out;
 508                         }
 509                         break;
 510
 511                 case T_TSSFLT:
 512                         /*
 513                          * PSL_NT can be set in user mode and isn't cleared
 514                          * automatically when the kernel is entered.  This
 515                          * causes a TSS fault when the kernel attempts to
 516                          * `iret' because the TSS link is uninitialized.  We
 517                          * want to get this fault so that we can fix the
 518                          * problem here and not every time the kernel is
 519                          * entered.
 520                          */
 521                         if (frame->tf_rflags & PSL_NT) {
 522                                 frame->tf_rflags &= ~PSL_NT;
 523                                 goto out;
 524                         }
 525                         break;
 526
 527                 case T_TRCTRAP:  /* trace trap */
 528                         /*
 529                          * Ignore debug register trace traps due to
 530                          * accesses in the user's address space, which
 531                          * can happen under several conditions such as
 532                          * if a user sets a watchpoint on a buffer and
 533                          * then passes that buffer to a system call.
 534                          * We still want to get TRCTRAPS for addresses
 535                          * in kernel space because that is useful when
 536                          * debugging the kernel.
 537                          */
 538                         if (user_dbreg_trap()) {
 539                                 /*
 540                                  * Reset breakpoint bits because the
 541                                  * processor doesn't
 542                                  */
 543                                 /* XXX check upper bits here */
 544                                 load_dr6(rdr6() & 0xfffffff0);
 545                                 goto out;
 546                         }
 547                         /*
 548                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 549                          */
 550                 case T_BPTFLT:
 551                         /*
 552                          * If KDB is enabled, let it handle the debugger trap.
 553                          * Otherwise, debugger traps "can't happen".
 554                          */
 555 #ifdef KDB
 556                         if (kdb_trap(type, 0, frame))
 557                                 goto out;
 558 #endif
 559                         break;
 560
 561 #ifdef DEV_ISA
 562                 case T_NMI:
 563                         /* machine/parity/power fail/"kitchen sink" faults */
 564                         if (isa_nmi(code) == 0) {
 565 #ifdef KDB
 566                                 /*
 567                                  * NMI can be hooked up to a pushbutton
 568                                  * for debugging.
 569                                  */
 570                                 if (kdb_on_nmi) {
 571                                         printf ("NMI ... going to debugger\n");
 572                                         kdb_trap(type, 0, frame);
 573                                 }
 574 #endif /* KDB */
 575                                 goto out;
 576                         } else if (panic_on_nmi == 0)
 577                                 goto out;
 578                         /* FALLTHROUGH */
 579 #endif /* DEV_ISA */
 580                 }
 581
 582                 trap_fatal(frame, 0);
 583                 goto out;
 584         }
 585
 586         /* Translate fault for emulators (e.g. Linux) */
 587         if (*p->p_sysent->sv_transtrap)
 588                 i = (*p->p_sysent->sv_transtrap)(i, type);
 589
 590         ksiginfo_init_trap(&ksi);
 591         ksi.ksi_signo = i;
 592         ksi.ksi_code = ucode;
 593         ksi.ksi_trapno = type;
 594         ksi.ksi_addr = (void *)addr;
 595         if (uprintf_signal) {
 596                 uprintf("pid %d comm %s: signal %d err %lx code %d type %d "
 597                     "addr 0x%lx rsp 0x%lx rip 0x%lx "
 598                     "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
 599                     p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr,
 600                     frame->tf_rsp, frame->tf_rip,
 601                     fubyte((void *)(frame->tf_rip + 0)),
 602                     fubyte((void *)(frame->tf_rip + 1)),
 603                     fubyte((void *)(frame->tf_rip + 2)),
 604                     fubyte((void *)(frame->tf_rip + 3)),
 605                     fubyte((void *)(frame->tf_rip + 4)),
 606                     fubyte((void *)(frame->tf_rip + 5)),
 607                     fubyte((void *)(frame->tf_rip + 6)),
 608                     fubyte((void *)(frame->tf_rip + 7)));
 609         }
 610         KASSERT((read_rflags() & PSL_I) != 0, ("interrupts disabled"));
 611         trapsignal(td, &ksi);
 612
 613 user:
 614         userret(td, frame);
 615         KASSERT(PCB_USER_FPU(td->td_pcb),
 616             ("Return from trap with kernel FPU ctx leaked"));
 617 userout:
 618 out:
 619         return;
 620 }
 621
 622 static int
 623 trap_pfault(frame, usermode)
 624         struct trapframe *frame;
 625         int usermode;
 626 {
 627         vm_offset_t va;
 628         struct vmspace *vm;
 629         vm_map_t map;
 630         int rv = 0;
 631         vm_prot_t ftype;
 632         struct thread *td = curthread;
 633         struct proc *p = td->td_proc;
 634         vm_offset_t eva = frame->tf_addr;
 635
 636         if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
 637                 /*
 638                  * Due to both processor errata and lazy TLB invalidation when
 639                  * access restrictions are removed from virtual pages, memory
 640                  * accesses that are allowed by the physical mapping layer may
 641                  * nonetheless cause one spurious page fault per virtual page.
 642                  * When the thread is executing a "no faulting" section that
 643                  * is bracketed by vm_fault_{disable,enable}_pagefaults(),
 644                  * every page fault is treated as a spurious page fault,
 645                  * unless it accesses the same virtual address as the most
 646                  * recent page fault within the same "no faulting" section.
 647                  */
 648                 if (td->td_md.md_spurflt_addr != eva ||
 649                     (td->td_pflags & TDP_RESETSPUR) != 0) {
 650                         /*
 651                          * Do nothing to the TLB.  A stale TLB entry is
 652                          * flushed automatically by a page fault.
 653                          */
 654                         td->td_md.md_spurflt_addr = eva;
 655                         td->td_pflags &= ~TDP_RESETSPUR;
 656                         return (0);
 657                 }
 658         } else {
 659                 /*
 660                  * If we get a page fault while in a critical section, then
 661                  * it is most likely a fatal kernel page fault.  The kernel
 662                  * is already going to panic trying to get a sleep lock to
 663                  * do the VM lookup, so just consider it a fatal trap so the
 664                  * kernel can print out a useful trap message and even get
 665                  * to the debugger.
 666                  *
 667                  * If we get a page fault while holding a non-sleepable
 668                  * lock, then it is most likely a fatal kernel page fault.
 669                  * If WITNESS is enabled, then it's going to whine about
 670                  * bogus LORs with various VM locks, so just skip to the
 671                  * fatal trap handling directly.
 672                  */
 673                 if (td->td_critnest != 0 ||
 674                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 675                     "Kernel page fault") != 0) {
 676                         trap_fatal(frame, eva);
 677                         return (-1);
 678                 }
 679         }
 680         va = trunc_page(eva);
 681         if (va >= VM_MIN_KERNEL_ADDRESS) {
 682                 /*
 683                  * Don't allow user-mode faults in kernel address space.
 684                  */
 685                 if (usermode)
 686                         goto nogo;
 687
 688                 map = kernel_map;
 689         } else {
 690                 /*
 691                  * This is a fault on non-kernel virtual memory.  If either
 692                  * p or p->p_vmspace is NULL, then the fault is fatal.
 693                  */
 694                 if (p == NULL || (vm = p->p_vmspace) == NULL)
 695                         goto nogo;
 696
 697                 map = &vm->vm_map;
 698
 699                 /*
 700                  * When accessing a usermode address, kernel must be
 701                  * ready to accept the page fault, and provide a
 702                  * handling routine.  Since accessing the address
 703                  * without the handler is a bug, do not try to handle
 704                  * it normally, and panic immediately.
 705                  */
 706                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 707                     curpcb->pcb_onfault == NULL)) {
 708                         trap_fatal(frame, eva);
 709                         return (-1);
 710                 }
 711         }
 712
 713         /*
 714          * If the trap was caused by errant bits in the PTE then panic.
 715          */
 716         if (frame->tf_err & PGEX_RSV) {
 717                 trap_fatal(frame, eva);
 718                 return (-1);
 719         }
 720
 721         /*
 722          * PGEX_I is defined only if the execute disable bit capability is
 723          * supported and enabled.
 724          */
 725         if (frame->tf_err & PGEX_W)
 726                 ftype = VM_PROT_WRITE;
 727         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 728                 ftype = VM_PROT_EXECUTE;
 729         else
 730                 ftype = VM_PROT_READ;
 731
 732         if (map != kernel_map) {
 733                 /*
 734                  * Keep swapout from messing with us during this
 735                  *      critical time.
 736                  */
 737                 PROC_LOCK(p);
 738                 ++p->p_lock;
 739                 PROC_UNLOCK(p);
 740
 741                 /* Fault in the user page: */
 742                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 743
 744                 PROC_LOCK(p);
 745                 --p->p_lock;
 746                 PROC_UNLOCK(p);
 747         } else {
 748                 /*
 749                  * Don't have to worry about process locking or stacks in the
 750                  * kernel.
 751                  */
 752                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 753         }
 754         if (rv == KERN_SUCCESS) {
 755 #ifdef HWPMC_HOOKS
 756                 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
 757                         PMC_SOFT_CALL_TF( , , page_fault, all, frame);
 758                         if (ftype == VM_PROT_READ)
 759                                 PMC_SOFT_CALL_TF( , , page_fault, read,
 760                                     frame);
 761                         else
 762                                 PMC_SOFT_CALL_TF( , , page_fault, write,
 763                                     frame);
 764                 }
 765 #endif
 766                 return (0);
 767         }
 768 nogo:
 769         if (!usermode) {
 770                 if (td->td_intr_nesting_level == 0 &&
 771                     curpcb->pcb_onfault != NULL) {
 772                         frame->tf_rip = (long)curpcb->pcb_onfault;
 773                         return (0);
 774                 }
 775                 if ((td->td_pflags & TDP_DEVMEMIO) != 0) {
 776                         KASSERT(curpcb->pcb_onfault != NULL,
 777                             ("/dev/mem without pcb_onfault"));
 778                         frame->tf_rip = (long)curpcb->pcb_onfault;
 779                         return (0);
 780                 }
 781                 trap_fatal(frame, eva);
 782                 return (-1);
 783         }
 784         return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 785 }
 786
 787 static void
 788 trap_fatal(frame, eva)
 789         struct trapframe *frame;
 790         vm_offset_t eva;
 791 {
 792         int code, ss;
 793         u_int type;
 794         long esp;
 795         struct soft_segment_descriptor softseg;
 796         char *msg;
 797
 798         code = frame->tf_err;
 799         type = frame->tf_trapno;
 800         sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)],
 801             &softseg);
 802
 803         if (type <= MAX_TRAP_MSG)
 804                 msg = trap_msg[type];
 805         else
 806                 msg = "UNKNOWN";
 807         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 808             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
 809 #ifdef SMP
 810         /* two separate prints in case of a trap on an unmapped page */
 811         printf("cpuid = %d; ", PCPU_GET(cpuid));
 812         printf("apic id = %02x\n", PCPU_GET(apic_id));
 813 #endif
 814         if (type == T_PAGEFLT) {
 815                 printf("fault virtual address   = 0x%lx\n", eva);
 816                 printf("fault code              = %s %s %s%s, %s\n",
 817                         code & PGEX_U ? "user" : "supervisor",
 818                         code & PGEX_W ? "write" : "read",
 819                         code & PGEX_I ? "instruction" : "data",
 820                         code & PGEX_RSV ? " rsv" : "",
 821                         code & PGEX_P ? "protection violation" : "page not present");
 822         }
 823         printf("instruction pointer     = 0x%lx:0x%lx\n",
 824                frame->tf_cs & 0xffff, frame->tf_rip);
 825         if (ISPL(frame->tf_cs) == SEL_UPL) {
 826                 ss = frame->tf_ss & 0xffff;
 827                 esp = frame->tf_rsp;
 828         } else {
 829                 ss = GSEL(GDATA_SEL, SEL_KPL);
 830                 esp = (long)&frame->tf_rsp;
 831         }
 832         printf("stack pointer           = 0x%x:0x%lx\n", ss, esp);
 833         printf("frame pointer           = 0x%x:0x%lx\n", ss, frame->tf_rbp);
 834         printf("code segment            = base 0x%lx, limit 0x%lx, type 0x%x\n",
 835                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 836         printf("                        = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
 837                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
 838                softseg.ssd_gran);
 839         printf("processor eflags        = ");
 840         if (frame->tf_rflags & PSL_T)
 841                 printf("trace trap, ");
 842         if (frame->tf_rflags & PSL_I)
 843                 printf("interrupt enabled, ");
 844         if (frame->tf_rflags & PSL_NT)
 845                 printf("nested task, ");
 846         if (frame->tf_rflags & PSL_RF)
 847                 printf("resume, ");
 848         printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
 849         printf("current process         = ");
 850         if (curproc) {
 851                 printf("%lu (%s)\n",
 852                     (u_long)curproc->p_pid, curthread->td_name ?
 853                     curthread->td_name : "");
 854         } else {
 855                 printf("Idle\n");
 856         }
 857
 858 #ifdef KDB
 859         if (debugger_on_panic || kdb_active)
 860                 if (kdb_trap(type, 0, frame))
 861                         return;
 862 #endif
 863         printf("trap number             = %d\n", type);
 864         if (type <= MAX_TRAP_MSG)
 865                 panic("%s", trap_msg[type]);
 866         else
 867                 panic("unknown/reserved trap");
 868 }
 869
 870 /*
 871  * Double fault handler. Called when a fault occurs while writing
 872  * a frame for a trap/exception onto the stack. This usually occurs
 873  * when the stack overflows (such is the case with infinite recursion,
 874  * for example).
 875  */
 876 void
 877 dblfault_handler(struct trapframe *frame)
 878 {
 879 #ifdef KDTRACE_HOOKS
 880         if (dtrace_doubletrap_func != NULL)
 881                 (*dtrace_doubletrap_func)();
 882 #endif
 883         printf("\nFatal double fault\n");
 884         printf("rip = 0x%lx\n", frame->tf_rip);
 885         printf("rsp = 0x%lx\n", frame->tf_rsp);
 886         printf("rbp = 0x%lx\n", frame->tf_rbp);
 887 #ifdef SMP
 888         /* two separate prints in case of a trap on an unmapped page */
 889         printf("cpuid = %d; ", PCPU_GET(cpuid));
 890         printf("apic id = %02x\n", PCPU_GET(apic_id));
 891 #endif
 892         panic("double fault");
 893 }
 894
 895 int
 896 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
 897 {
 898         struct proc *p;
 899         struct trapframe *frame;
 900         register_t *argp;
 901         caddr_t params;
 902         int reg, regcnt, error;
 903
 904         p = td->td_proc;
 905         frame = td->td_frame;
 906         reg = 0;
 907         regcnt = 6;
 908
 909         params = (caddr_t)frame->tf_rsp + sizeof(register_t);
 910         sa->code = frame->tf_rax;
 911
 912         if (sa->code == SYS_syscall || sa->code == SYS___syscall) {
 913                 sa->code = frame->tf_rdi;
 914                 reg++;
 915                 regcnt--;
 916         }
 917         if (p->p_sysent->sv_mask)
 918                 sa->code &= p->p_sysent->sv_mask;
 919
 920         if (sa->code >= p->p_sysent->sv_size)
 921                 sa->callp = &p->p_sysent->sv_table[0];
 922         else
 923                 sa->callp = &p->p_sysent->sv_table[sa->code];
 924
 925         sa->narg = sa->callp->sy_narg;
 926         KASSERT(sa->narg <= sizeof(sa->args) / sizeof(sa->args[0]),
 927             ("Too many syscall arguments!"));
 928         error = 0;
 929         argp = &frame->tf_rdi;
 930         argp += reg;
 931         bcopy(argp, sa->args, sizeof(sa->args[0]) * regcnt);
 932         if (sa->narg > regcnt) {
 933                 KASSERT(params != NULL, ("copyin args with no params!"));
 934                 error = copyin(params, &sa->args[regcnt],
 935                     (sa->narg - regcnt) * sizeof(sa->args[0]));
 936         }
 937
 938         if (error == 0) {
 939                 td->td_retval[0] = 0;
 940                 td->td_retval[1] = frame->tf_rdx;
 941         }
 942
 943         return (error);
 944 }
 945
 946 #include "../../kern/subr_syscall.c"
 947
 948 /*
 949  * System call handler for native binaries.  The trap frame is already
 950  * set up by the assembler trampoline and a pointer to it is saved in
 951  * td_frame.
 952  */
 953 void
 954 amd64_syscall(struct thread *td, int traced)
 955 {
 956         struct syscall_args sa;
 957         int error;
 958         ksiginfo_t ksi;
 959
 960 #ifdef DIAGNOSTIC
 961         if (ISPL(td->td_frame->tf_cs) != SEL_UPL) {
 962                 panic("syscall");
 963                 /* NOT REACHED */
 964         }
 965 #endif
 966         error = syscallenter(td, &sa);
 967
 968         /*
 969          * Traced syscall.
 970          */
 971         if (__predict_false(traced)) {
 972                 td->td_frame->tf_rflags &= ~PSL_T;
 973                 ksiginfo_init_trap(&ksi);
 974                 ksi.ksi_signo = SIGTRAP;
 975                 ksi.ksi_code = TRAP_TRACE;
 976                 ksi.ksi_addr = (void *)td->td_frame->tf_rip;
 977                 trapsignal(td, &ksi);
 978         }
 979
 980         KASSERT(PCB_USER_FPU(td->td_pcb),
 981             ("System call %s returing with kernel FPU ctx leaked",
 982              syscallname(td->td_proc, sa.code)));
 983         KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td),
 984             ("System call %s returning with mangled pcb_save",
 985              syscallname(td->td_proc, sa.code)));
 986
 987         syscallret(td, error, &sa);
 988
 989         /*
 990          * If the user-supplied value of %rip is not a canonical
 991          * address, then some CPUs will trigger a ring 0 #GP during
 992          * the sysret instruction.  However, the fault handler would
 993          * execute in ring 0 with the user's %gs and %rsp which would
 994          * not be safe.  Instead, use the full return path which
 995          * catches the problem safely.
 996          */
 997         if (td->td_frame->tf_rip >= VM_MAXUSER_ADDRESS)
 998                 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
 999 }