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
 100 /*
 101  * This is a hook which is initialised by the dtrace module
 102  * to handle traps which might occur during DTrace probe
 103  * execution.
 104  */
 105 dtrace_trap_func_t      dtrace_trap_func;
 106
 107 dtrace_doubletrap_func_t        dtrace_doubletrap_func;
 108
 109 /*
 110  * This is a hook which is initialised by the systrace module
 111  * when it is loaded. This keeps the DTrace syscall provider
 112  * implementation opaque.
 113  */
 114 systrace_probe_func_t   systrace_probe_func;
 115
 116 /*
 117  * These hooks are necessary for the pid and usdt providers.
 118  */
 119 dtrace_pid_probe_ptr_t          dtrace_pid_probe_ptr;
 120 dtrace_return_probe_ptr_t       dtrace_return_probe_ptr;
 121 #endif
 122
 123 extern void trap(struct trapframe *frame);
 124 extern void syscall(struct trapframe *frame);
 125 void dblfault_handler(struct trapframe *frame);
 126
 127 static int trap_pfault(struct trapframe *, int);
 128 static void trap_fatal(struct trapframe *, vm_offset_t);
 129
 130 #define MAX_TRAP_MSG            32
 131 static char *trap_msg[] = {
 132         "",                                     /*  0 unused */
 133         "privileged instruction fault",         /*  1 T_PRIVINFLT */
 134         "",                                     /*  2 unused */
 135         "breakpoint instruction fault",         /*  3 T_BPTFLT */
 136         "",                                     /*  4 unused */
 137         "",                                     /*  5 unused */
 138         "arithmetic trap",                      /*  6 T_ARITHTRAP */
 139         "",                                     /*  7 unused */
 140         "",                                     /*  8 unused */
 141         "general protection fault",             /*  9 T_PROTFLT */
 142         "trace trap",                           /* 10 T_TRCTRAP */
 143         "",                                     /* 11 unused */
 144         "page fault",                           /* 12 T_PAGEFLT */
 145         "",                                     /* 13 unused */
 146         "alignment fault",                      /* 14 T_ALIGNFLT */
 147         "",                                     /* 15 unused */
 148         "",                                     /* 16 unused */
 149         "",                                     /* 17 unused */
 150         "integer divide fault",                 /* 18 T_DIVIDE */
 151         "non-maskable interrupt trap",          /* 19 T_NMI */
 152         "overflow trap",                        /* 20 T_OFLOW */
 153         "FPU bounds check fault",               /* 21 T_BOUND */
 154         "FPU device not available",             /* 22 T_DNA */
 155         "double fault",                         /* 23 T_DOUBLEFLT */
 156         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
 157         "invalid TSS fault",                    /* 25 T_TSSFLT */
 158         "segment not present fault",            /* 26 T_SEGNPFLT */
 159         "stack fault",                          /* 27 T_STKFLT */
 160         "machine check trap",                   /* 28 T_MCHK */
 161         "SIMD floating-point exception",        /* 29 T_XMMFLT */
 162         "reserved (unknown) fault",             /* 30 T_RESERVED */
 163         "",                                     /* 31 unused (reserved) */
 164         "DTrace pid return trap",               /* 32 T_DTRACE_RET */
 165 };
 166
 167 #ifdef KDB
 168 static int kdb_on_nmi = 1;
 169 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
 170         &kdb_on_nmi, 0, "Go to KDB on NMI");
 171 TUNABLE_INT("machdep.kdb_on_nmi", &kdb_on_nmi);
 172 #endif
 173 static int panic_on_nmi = 1;
 174 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
 175         &panic_on_nmi, 0, "Panic on NMI");
 176 TUNABLE_INT("machdep.panic_on_nmi", &panic_on_nmi);
 177 static int prot_fault_translation;
 178 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
 179     &prot_fault_translation, 0,
 180     "Select signal to deliver on protection fault");
 181 static int uprintf_signal;
 182 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW,
 183     &uprintf_signal, 0,
 184     "Print debugging information on trap signal to ctty");
 185
 186 /*
 187  * Exception, fault, and trap interface to the FreeBSD kernel.
 188  * This common code is called from assembly language IDT gate entry
 189  * routines that prepare a suitable stack frame, and restore this
 190  * frame after the exception has been processed.
 191  */
 192
 193 void
 194 trap(struct trapframe *frame)
 195 {
 196         struct thread *td = curthread;
 197         struct proc *p = td->td_proc;
 198         int i = 0, ucode = 0, code;
 199         u_int type;
 200         register_t addr = 0;
 201         ksiginfo_t ksi;
 202
 203         PCPU_INC(cnt.v_trap);
 204         type = frame->tf_trapno;
 205
 206 #ifdef SMP
 207         /* Handler for NMI IPIs used for stopping CPUs. */
 208         if (type == T_NMI) {
 209                  if (ipi_nmi_handler() == 0)
 210                            goto out;
 211         }
 212 #endif /* SMP */
 213
 214 #ifdef KDB
 215         if (kdb_active) {
 216                 kdb_reenter();
 217                 goto out;
 218         }
 219 #endif
 220
 221         if (type == T_RESERVED) {
 222                 trap_fatal(frame, 0);
 223                 goto out;
 224         }
 225
 226 #ifdef  HWPMC_HOOKS
 227         /*
 228          * CPU PMCs interrupt using an NMI.  If the PMC module is
 229          * active, pass the 'rip' value to the PMC module's interrupt
 230          * handler.  A return value of '1' from the handler means that
 231          * the NMI was handled by it and we can return immediately.
 232          */
 233         if (type == T_NMI && pmc_intr &&
 234             (*pmc_intr)(PCPU_GET(cpuid), frame))
 235                 goto out;
 236 #endif
 237
 238         if (type == T_MCHK) {
 239                 mca_intr();
 240                 goto out;
 241         }
 242
 243 #ifdef KDTRACE_HOOKS
 244         /*
 245          * A trap can occur while DTrace executes a probe. Before
 246          * executing the probe, DTrace blocks re-scheduling and sets
 247          * a flag in it's per-cpu flags to indicate that it doesn't
 248          * want to fault. On returning from the probe, the no-fault
 249          * flag is cleared and finally re-scheduling is enabled.
 250          *
 251          * If the DTrace kernel module has registered a trap handler,
 252          * call it and if it returns non-zero, assume that it has
 253          * handled the trap and modified the trap frame so that this
 254          * function can return normally.
 255          */
 256         if (type == T_DTRACE_RET || type == T_BPTFLT) {
 257                 struct reg regs;
 258
 259                 fill_frame_regs(frame, &regs);
 260                 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
 305         if (ISPL(frame->tf_cs) == SEL_UPL) {
 306                 /* user trap */
 307
 308                 td->td_pticks = 0;
 309                 td->td_frame = frame;
 310                 addr = frame->tf_rip;
 311                 if (td->td_ucred != p->p_ucred)
 312                         cred_update_thread(td);
 313
 314                 switch (type) {
 315                 case T_PRIVINFLT:       /* privileged instruction fault */
 316                         i = SIGILL;
 317                         ucode = ILL_PRVOPC;
 318                         break;
 319
 320                 case T_BPTFLT:          /* bpt instruction fault */
 321                 case T_TRCTRAP:         /* trace trap */
 322                         enable_intr();
 323                         frame->tf_rflags &= ~PSL_T;
 324                         i = SIGTRAP;
 325                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 326                         break;
 327
 328                 case T_ARITHTRAP:       /* arithmetic trap */
 329                         ucode = fputrap_x87();
 330                         if (ucode == -1)
 331                                 goto userout;
 332                         i = SIGFPE;
 333                         break;
 334
 335                 case T_PROTFLT:         /* general protection fault */
 336                         i = SIGBUS;
 337                         ucode = BUS_OBJERR;
 338                         break;
 339                 case T_STKFLT:          /* stack fault */
 340                 case T_SEGNPFLT:        /* segment not present fault */
 341                         i = SIGBUS;
 342                         ucode = BUS_ADRERR;
 343                         break;
 344                 case T_TSSFLT:          /* invalid TSS fault */
 345                         i = SIGBUS;
 346                         ucode = BUS_OBJERR;
 347                         break;
 348                 case T_DOUBLEFLT:       /* double fault */
 349                 default:
 350                         i = SIGBUS;
 351                         ucode = BUS_OBJERR;
 352                         break;
 353
 354                 case T_PAGEFLT:         /* page fault */
 355                         addr = frame->tf_addr;
 356                         i = trap_pfault(frame, TRUE);
 357                         if (i == -1)
 358                                 goto userout;
 359                         if (i == 0)
 360                                 goto user;
 361
 362                         if (i == SIGSEGV)
 363                                 ucode = SEGV_MAPERR;
 364                         else {
 365                                 if (prot_fault_translation == 0) {
 366                                         /*
 367                                          * Autodetect.
 368                                          * This check also covers the images
 369                                          * without the ABI-tag ELF note.
 370                                          */
 371                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 372                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 373                                                 i = SIGSEGV;
 374                                                 ucode = SEGV_ACCERR;
 375                                         } else {
 376                                                 i = SIGBUS;
 377                                                 ucode = BUS_PAGE_FAULT;
 378                                         }
 379                                 } else if (prot_fault_translation == 1) {
 380                                         /*
 381                                          * Always compat mode.
 382                                          */
 383                                         i = SIGBUS;
 384                                         ucode = BUS_PAGE_FAULT;
 385                                 } else {
 386                                         /*
 387                                          * Always SIGSEGV mode.
 388                                          */
 389                                         i = SIGSEGV;
 390                                         ucode = SEGV_ACCERR;
 391                                 }
 392                         }
 393                         break;
 394
 395                 case T_DIVIDE:          /* integer divide fault */
 396                         ucode = FPE_INTDIV;
 397                         i = SIGFPE;
 398                         break;
 399
 400 #ifdef DEV_ISA
 401                 case T_NMI:
 402                         /* machine/parity/power fail/"kitchen sink" faults */
 403                         if (isa_nmi(code) == 0) {
 404 #ifdef KDB
 405                                 /*
 406                                  * NMI can be hooked up to a pushbutton
 407                                  * for debugging.
 408                                  */
 409                                 if (kdb_on_nmi) {
 410                                         printf ("NMI ... going to debugger\n");
 411                                         kdb_trap(type, 0, frame);
 412                                 }
 413 #endif /* KDB */
 414                                 goto userout;
 415                         } else if (panic_on_nmi)
 416                                 panic("NMI indicates hardware failure");
 417                         break;
 418 #endif /* DEV_ISA */
 419
 420                 case T_OFLOW:           /* integer overflow fault */
 421                         ucode = FPE_INTOVF;
 422                         i = SIGFPE;
 423                         break;
 424
 425                 case T_BOUND:           /* bounds check fault */
 426                         ucode = FPE_FLTSUB;
 427                         i = SIGFPE;
 428                         break;
 429
 430                 case T_DNA:
 431                         /* transparent fault (due to context switch "late") */
 432                         KASSERT(PCB_USER_FPU(td->td_pcb),
 433                             ("kernel FPU ctx has leaked"));
 434                         fpudna();
 435                         goto userout;
 436
 437                 case T_FPOPFLT:         /* FPU operand fetch fault */
 438                         ucode = ILL_COPROC;
 439                         i = SIGILL;
 440                         break;
 441
 442                 case T_XMMFLT:          /* SIMD floating-point exception */
 443                         ucode = fputrap_sse();
 444                         if (ucode == -1)
 445                                 goto userout;
 446                         i = SIGFPE;
 447                         break;
 448                 }
 449         } else {
 450                 /* kernel trap */
 451
 452                 KASSERT(cold || td->td_ucred != NULL,
 453                     ("kernel trap doesn't have ucred"));
 454                 switch (type) {
 455                 case T_PAGEFLT:                 /* page fault */
 456                         (void) trap_pfault(frame, FALSE);
 457                         goto out;
 458
 459                 case T_DNA:
 460                         KASSERT(!PCB_USER_FPU(td->td_pcb),
 461                             ("Unregistered use of FPU in kernel"));
 462                         fpudna();
 463                         goto out;
 464
 465                 case T_ARITHTRAP:       /* arithmetic trap */
 466                 case T_XMMFLT:          /* SIMD floating-point exception */
 467                 case T_FPOPFLT:         /* FPU operand fetch fault */
 468                         /*
 469                          * XXXKIB for now disable any FPU traps in kernel
 470                          * handler registration seems to be overkill
 471                          */
 472                         trap_fatal(frame, 0);
 473                         goto out;
 474
 475                 case T_STKFLT:          /* stack fault */
 476                         break;
 477
 478                 case T_PROTFLT:         /* general protection fault */
 479                 case T_SEGNPFLT:        /* segment not present fault */
 480                         if (td->td_intr_nesting_level != 0)
 481                                 break;
 482
 483                         /*
 484                          * Invalid segment selectors and out of bounds
 485                          * %rip's and %rsp's can be set up in user mode.
 486                          * This causes a fault in kernel mode when the
 487                          * kernel tries to return to user mode.  We want
 488                          * to get this fault so that we can fix the
 489                          * problem here and not have to check all the
 490                          * selectors and pointers when the user changes
 491                          * them.
 492                          */
 493                         if (frame->tf_rip == (long)doreti_iret) {
 494                                 frame->tf_rip = (long)doreti_iret_fault;
 495                                 goto out;
 496                         }
 497                         if (frame->tf_rip == (long)ld_ds) {
 498                                 frame->tf_rip = (long)ds_load_fault;
 499                                 goto out;
 500                         }
 501                         if (frame->tf_rip == (long)ld_es) {
 502                                 frame->tf_rip = (long)es_load_fault;
 503                                 goto out;
 504                         }
 505                         if (frame->tf_rip == (long)ld_fs) {
 506                                 frame->tf_rip = (long)fs_load_fault;
 507                                 goto out;
 508                         }
 509                         if (frame->tf_rip == (long)ld_gs) {
 510                                 frame->tf_rip = (long)gs_load_fault;
 511                                 goto out;
 512                         }
 513                         if (frame->tf_rip == (long)ld_gsbase) {
 514                                 frame->tf_rip = (long)gsbase_load_fault;
 515                                 goto out;
 516                         }
 517                         if (frame->tf_rip == (long)ld_fsbase) {
 518                                 frame->tf_rip = (long)fsbase_load_fault;
 519                                 goto out;
 520                         }
 521                         if (curpcb->pcb_onfault != NULL) {
 522                                 frame->tf_rip = (long)curpcb->pcb_onfault;
 523                                 goto out;
 524                         }
 525                         break;
 526
 527                 case T_TSSFLT:
 528                         /*
 529                          * PSL_NT can be set in user mode and isn't cleared
 530                          * automatically when the kernel is entered.  This
 531                          * causes a TSS fault when the kernel attempts to
 532                          * `iret' because the TSS link is uninitialized.  We
 533                          * want to get this fault so that we can fix the
 534                          * problem here and not every time the kernel is
 535                          * entered.
 536                          */
 537                         if (frame->tf_rflags & PSL_NT) {
 538                                 frame->tf_rflags &= ~PSL_NT;
 539                                 goto out;
 540                         }
 541                         break;
 542
 543                 case T_TRCTRAP:  /* trace trap */
 544                         /*
 545                          * Ignore debug register trace traps due to
 546                          * accesses in the user's address space, which
 547                          * can happen under several conditions such as
 548                          * if a user sets a watchpoint on a buffer and
 549                          * then passes that buffer to a system call.
 550                          * We still want to get TRCTRAPS for addresses
 551                          * in kernel space because that is useful when
 552                          * debugging the kernel.
 553                          */
 554                         if (user_dbreg_trap()) {
 555                                 /*
 556                                  * Reset breakpoint bits because the
 557                                  * processor doesn't
 558                                  */
 559                                 /* XXX check upper bits here */
 560                                 load_dr6(rdr6() & 0xfffffff0);
 561                                 goto out;
 562                         }
 563                         /*
 564                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 565                          */
 566                 case T_BPTFLT:
 567                         /*
 568                          * If KDB is enabled, let it handle the debugger trap.
 569                          * Otherwise, debugger traps "can't happen".
 570                          */
 571 #ifdef KDB
 572                         if (kdb_trap(type, 0, frame))
 573                                 goto out;
 574 #endif
 575                         break;
 576
 577 #ifdef DEV_ISA
 578                 case T_NMI:
 579                         /* machine/parity/power fail/"kitchen sink" faults */
 580                         if (isa_nmi(code) == 0) {
 581 #ifdef KDB
 582                                 /*
 583                                  * NMI can be hooked up to a pushbutton
 584                                  * for debugging.
 585                                  */
 586                                 if (kdb_on_nmi) {
 587                                         printf ("NMI ... going to debugger\n");
 588                                         kdb_trap(type, 0, frame);
 589                                 }
 590 #endif /* KDB */
 591                                 goto out;
 592                         } else if (panic_on_nmi == 0)
 593                                 goto out;
 594                         /* FALLTHROUGH */
 595 #endif /* DEV_ISA */
 596                 }
 597
 598                 trap_fatal(frame, 0);
 599                 goto out;
 600         }
 601
 602         /* Translate fault for emulators (e.g. Linux) */
 603         if (*p->p_sysent->sv_transtrap)
 604                 i = (*p->p_sysent->sv_transtrap)(i, type);
 605
 606         ksiginfo_init_trap(&ksi);
 607         ksi.ksi_signo = i;
 608         ksi.ksi_code = ucode;
 609         ksi.ksi_trapno = type;
 610         ksi.ksi_addr = (void *)addr;
 611         if (uprintf_signal) {
 612                 uprintf("pid %d comm %s: signal %d err %lx code %d type %d "
 613                     "addr 0x%lx rsp 0x%lx rip 0x%lx "
 614                     "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
 615                     p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr,
 616                     frame->tf_rsp, frame->tf_rip,
 617                     fubyte((void *)(frame->tf_rip + 0)),
 618                     fubyte((void *)(frame->tf_rip + 1)),
 619                     fubyte((void *)(frame->tf_rip + 2)),
 620                     fubyte((void *)(frame->tf_rip + 3)),
 621                     fubyte((void *)(frame->tf_rip + 4)),
 622                     fubyte((void *)(frame->tf_rip + 5)),
 623                     fubyte((void *)(frame->tf_rip + 6)),
 624                     fubyte((void *)(frame->tf_rip + 7)));
 625         }
 626         KASSERT((read_rflags() & PSL_I) != 0, ("interrupts disabled"));
 627         trapsignal(td, &ksi);
 628
 629 user:
 630         userret(td, frame);
 631         mtx_assert(&Giant, MA_NOTOWNED);
 632         KASSERT(PCB_USER_FPU(td->td_pcb),
 633             ("Return from trap with kernel FPU ctx leaked"));
 634 userout:
 635 out:
 636         return;
 637 }
 638
 639 static int
 640 trap_pfault(frame, usermode)
 641         struct trapframe *frame;
 642         int usermode;
 643 {
 644         vm_offset_t va;
 645         struct vmspace *vm = NULL;
 646         vm_map_t map;
 647         int rv = 0;
 648         vm_prot_t ftype;
 649         struct thread *td = curthread;
 650         struct proc *p = td->td_proc;
 651         vm_offset_t eva = frame->tf_addr;
 652
 653         if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
 654                 /*
 655                  * Due to both processor errata and lazy TLB invalidation when
 656                  * access restrictions are removed from virtual pages, memory
 657                  * accesses that are allowed by the physical mapping layer may
 658                  * nonetheless cause one spurious page fault per virtual page.
 659                  * When the thread is executing a "no faulting" section that
 660                  * is bracketed by vm_fault_{disable,enable}_pagefaults(),
 661                  * every page fault is treated as a spurious page fault,
 662                  * unless it accesses the same virtual address as the most
 663                  * recent page fault within the same "no faulting" section.
 664                  */
 665                 if (td->td_md.md_spurflt_addr != eva ||
 666                     (td->td_pflags & TDP_RESETSPUR) != 0) {
 667                         /*
 668                          * Do nothing to the TLB.  A stale TLB entry is
 669                          * flushed automatically by a page fault.
 670                          */
 671                         td->td_md.md_spurflt_addr = eva;
 672                         td->td_pflags &= ~TDP_RESETSPUR;
 673                         return (0);
 674                 }
 675         } else {
 676                 /*
 677                  * If we get a page fault while in a critical section, then
 678                  * it is most likely a fatal kernel page fault.  The kernel
 679                  * is already going to panic trying to get a sleep lock to
 680                  * do the VM lookup, so just consider it a fatal trap so the
 681                  * kernel can print out a useful trap message and even get
 682                  * to the debugger.
 683                  *
 684                  * If we get a page fault while holding a non-sleepable
 685                  * lock, then it is most likely a fatal kernel page fault.
 686                  * If WITNESS is enabled, then it's going to whine about
 687                  * bogus LORs with various VM locks, so just skip to the
 688                  * fatal trap handling directly.
 689                  */
 690                 if (td->td_critnest != 0 ||
 691                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 692                     "Kernel page fault") != 0) {
 693                         trap_fatal(frame, eva);
 694                         return (-1);
 695                 }
 696         }
 697         va = trunc_page(eva);
 698         if (va >= VM_MIN_KERNEL_ADDRESS) {
 699                 /*
 700                  * Don't allow user-mode faults in kernel address space.
 701                  */
 702                 if (usermode)
 703                         goto nogo;
 704
 705                 map = kernel_map;
 706         } else {
 707                 /*
 708                  * This is a fault on non-kernel virtual memory.
 709                  * vm is initialized above to NULL. If curproc is NULL
 710                  * or curproc->p_vmspace is NULL the fault is fatal.
 711                  */
 712                 if (p != NULL)
 713                         vm = p->p_vmspace;
 714
 715                 if (vm == NULL)
 716                         goto nogo;
 717
 718                 map = &vm->vm_map;
 719
 720                 /*
 721                  * When accessing a usermode address, kernel must be
 722                  * ready to accept the page fault, and provide a
 723                  * handling routine.  Since accessing the address
 724                  * without the handler is a bug, do not try to handle
 725                  * it normally, and panic immediately.
 726                  */
 727                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 728                     curpcb->pcb_onfault == NULL)) {
 729                         trap_fatal(frame, eva);
 730                         return (-1);
 731                 }
 732         }
 733
 734         /*
 735          * PGEX_I is defined only if the execute disable bit capability is
 736          * supported and enabled.
 737          */
 738         if (frame->tf_err & PGEX_W)
 739                 ftype = VM_PROT_WRITE;
 740         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 741                 ftype = VM_PROT_EXECUTE;
 742         else
 743                 ftype = VM_PROT_READ;
 744
 745         if (map != kernel_map) {
 746                 /*
 747                  * Keep swapout from messing with us during this
 748                  *      critical time.
 749                  */
 750                 PROC_LOCK(p);
 751                 ++p->p_lock;
 752                 PROC_UNLOCK(p);
 753
 754                 /* Fault in the user page: */
 755                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 756
 757                 PROC_LOCK(p);
 758                 --p->p_lock;
 759                 PROC_UNLOCK(p);
 760         } else {
 761                 /*
 762                  * Don't have to worry about process locking or stacks in the
 763                  * kernel.
 764                  */
 765                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 766         }
 767         if (rv == KERN_SUCCESS) {
 768 #ifdef HWPMC_HOOKS
 769                 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
 770                         PMC_SOFT_CALL_TF( , , page_fault, all, frame);
 771                         if (ftype == VM_PROT_READ)
 772                                 PMC_SOFT_CALL_TF( , , page_fault, read,
 773                                     frame);
 774                         else
 775                                 PMC_SOFT_CALL_TF( , , page_fault, write,
 776                                     frame);
 777                 }
 778 #endif
 779                 return (0);
 780         }
 781 nogo:
 782         if (!usermode) {
 783                 if (td->td_intr_nesting_level == 0 &&
 784                     curpcb->pcb_onfault != NULL) {
 785                         frame->tf_rip = (long)curpcb->pcb_onfault;
 786                         return (0);
 787                 }
 788                 if ((td->td_pflags & TDP_DEVMEMIO) != 0) {
 789                         KASSERT(curpcb->pcb_onfault != NULL,
 790                             ("/dev/mem without pcb_onfault"));
 791                         frame->tf_rip = (long)curpcb->pcb_onfault;
 792                         return (0);
 793                 }
 794                 trap_fatal(frame, eva);
 795                 return (-1);
 796         }
 797
 798         return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 799 }
 800
 801 static void
 802 trap_fatal(frame, eva)
 803         struct trapframe *frame;
 804         vm_offset_t eva;
 805 {
 806         int code, ss;
 807         u_int type;
 808         long esp;
 809         struct soft_segment_descriptor softseg;
 810         char *msg;
 811
 812         code = frame->tf_err;
 813         type = frame->tf_trapno;
 814         sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)],
 815             &softseg);
 816
 817         if (type <= MAX_TRAP_MSG)
 818                 msg = trap_msg[type];
 819         else
 820                 msg = "UNKNOWN";
 821         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 822             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
 823 #ifdef SMP
 824         /* two separate prints in case of a trap on an unmapped page */
 825         printf("cpuid = %d; ", PCPU_GET(cpuid));
 826         printf("apic id = %02x\n", PCPU_GET(apic_id));
 827 #endif
 828         if (type == T_PAGEFLT) {
 829                 printf("fault virtual address   = 0x%lx\n", eva);
 830                 printf("fault code              = %s %s %s, %s\n",
 831                         code & PGEX_U ? "user" : "supervisor",
 832                         code & PGEX_W ? "write" : "read",
 833                         code & PGEX_I ? "instruction" : "data",
 834                         code & PGEX_P ? "protection violation" : "page not present");
 835         }
 836         printf("instruction pointer     = 0x%lx:0x%lx\n",
 837                frame->tf_cs & 0xffff, frame->tf_rip);
 838         if (ISPL(frame->tf_cs) == SEL_UPL) {
 839                 ss = frame->tf_ss & 0xffff;
 840                 esp = frame->tf_rsp;
 841         } else {
 842                 ss = GSEL(GDATA_SEL, SEL_KPL);
 843                 esp = (long)&frame->tf_rsp;
 844         }
 845         printf("stack pointer           = 0x%x:0x%lx\n", ss, esp);
 846         printf("frame pointer           = 0x%x:0x%lx\n", ss, frame->tf_rbp);
 847         printf("code segment            = base 0x%lx, limit 0x%lx, type 0x%x\n",
 848                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 849         printf("                        = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
 850                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
 851                softseg.ssd_gran);
 852         printf("processor eflags        = ");
 853         if (frame->tf_rflags & PSL_T)
 854                 printf("trace trap, ");
 855         if (frame->tf_rflags & PSL_I)
 856                 printf("interrupt enabled, ");
 857         if (frame->tf_rflags & PSL_NT)
 858                 printf("nested task, ");
 859         if (frame->tf_rflags & PSL_RF)
 860                 printf("resume, ");
 861         printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
 862         printf("current process         = ");
 863         if (curproc) {
 864                 printf("%lu (%s)\n",
 865                     (u_long)curproc->p_pid, curthread->td_name ?
 866                     curthread->td_name : "");
 867         } else {
 868                 printf("Idle\n");
 869         }
 870
 871 #ifdef KDB
 872         if (debugger_on_panic || kdb_active)
 873                 if (kdb_trap(type, 0, frame))
 874                         return;
 875 #endif
 876         printf("trap number             = %d\n", type);
 877         if (type <= MAX_TRAP_MSG)
 878                 panic("%s", trap_msg[type]);
 879         else
 880                 panic("unknown/reserved trap");
 881 }
 882
 883 /*
 884  * Double fault handler. Called when a fault occurs while writing
 885  * a frame for a trap/exception onto the stack. This usually occurs
 886  * when the stack overflows (such is the case with infinite recursion,
 887  * for example).
 888  */
 889 void
 890 dblfault_handler(struct trapframe *frame)
 891 {
 892 #ifdef KDTRACE_HOOKS
 893         if (dtrace_doubletrap_func != NULL)
 894                 (*dtrace_doubletrap_func)();
 895 #endif
 896         printf("\nFatal double fault\n");
 897         printf("rip = 0x%lx\n", frame->tf_rip);
 898         printf("rsp = 0x%lx\n", frame->tf_rsp);
 899         printf("rbp = 0x%lx\n", frame->tf_rbp);
 900 #ifdef SMP
 901         /* two separate prints in case of a trap on an unmapped page */
 902         printf("cpuid = %d; ", PCPU_GET(cpuid));
 903         printf("apic id = %02x\n", PCPU_GET(apic_id));
 904 #endif
 905         panic("double fault");
 906 }
 907
 908 int
 909 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
 910 {
 911         struct proc *p;
 912         struct trapframe *frame;
 913         register_t *argp;
 914         caddr_t params;
 915         int reg, regcnt, error;
 916
 917         p = td->td_proc;
 918         frame = td->td_frame;
 919         reg = 0;
 920         regcnt = 6;
 921
 922         params = (caddr_t)frame->tf_rsp + sizeof(register_t);
 923         sa->code = frame->tf_rax;
 924
 925         if (sa->code == SYS_syscall || sa->code == SYS___syscall) {
 926                 sa->code = frame->tf_rdi;
 927                 reg++;
 928                 regcnt--;
 929         }
 930         if (p->p_sysent->sv_mask)
 931                 sa->code &= p->p_sysent->sv_mask;
 932
 933         if (sa->code >= p->p_sysent->sv_size)
 934                 sa->callp = &p->p_sysent->sv_table[0];
 935         else
 936                 sa->callp = &p->p_sysent->sv_table[sa->code];
 937
 938         sa->narg = sa->callp->sy_narg;
 939         KASSERT(sa->narg <= sizeof(sa->args) / sizeof(sa->args[0]),
 940             ("Too many syscall arguments!"));
 941         error = 0;
 942         argp = &frame->tf_rdi;
 943         argp += reg;
 944         bcopy(argp, sa->args, sizeof(sa->args[0]) * regcnt);
 945         if (sa->narg > regcnt) {
 946                 KASSERT(params != NULL, ("copyin args with no params!"));
 947                 error = copyin(params, &sa->args[regcnt],
 948                     (sa->narg - regcnt) * sizeof(sa->args[0]));
 949         }
 950
 951         if (error == 0) {
 952                 td->td_retval[0] = 0;
 953                 td->td_retval[1] = frame->tf_rdx;
 954         }
 955
 956         return (error);
 957 }
 958
 959 #include "../../kern/subr_syscall.c"
 960
 961 /*
 962  * System call handler for native binaries.  The trap frame is already
 963  * set up by the assembler trampoline and a pointer to it is saved in
 964  * td_frame.
 965  */
 966 void
 967 amd64_syscall(struct thread *td, int traced)
 968 {
 969         struct syscall_args sa;
 970         int error;
 971         ksiginfo_t ksi;
 972
 973 #ifdef DIAGNOSTIC
 974         if (ISPL(td->td_frame->tf_cs) != SEL_UPL) {
 975                 panic("syscall");
 976                 /* NOT REACHED */
 977         }
 978 #endif
 979         error = syscallenter(td, &sa);
 980
 981         /*
 982          * Traced syscall.
 983          */
 984         if (__predict_false(traced)) {
 985                 td->td_frame->tf_rflags &= ~PSL_T;
 986                 ksiginfo_init_trap(&ksi);
 987                 ksi.ksi_signo = SIGTRAP;
 988                 ksi.ksi_code = TRAP_TRACE;
 989                 ksi.ksi_addr = (void *)td->td_frame->tf_rip;
 990                 trapsignal(td, &ksi);
 991         }
 992
 993         KASSERT(PCB_USER_FPU(td->td_pcb),
 994             ("System call %s returing with kernel FPU ctx leaked",
 995              syscallname(td->td_proc, sa.code)));
 996         KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td),
 997             ("System call %s returning with mangled pcb_save",
 998              syscallname(td->td_proc, sa.code)));
 999
1000         syscallret(td, error, &sa);
1001
1002         /*
1003          * If the user-supplied value of %rip is not a canonical
1004          * address, then some CPUs will trigger a ring 0 #GP during
1005          * the sysret instruction.  However, the fault handler would
1006          * execute in ring 0 with the user's %gs and %rsp which would
1007          * not be safe.  Instead, use the full return path which
1008          * catches the problem safely.
1009          */
1010         if (td->td_frame->tf_rip >= VM_MAXUSER_ADDRESS)
1011                 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
1012 }