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_stack.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/stack.h>
  96 #include <machine/tss.h>
  97
  98 #ifdef KDTRACE_HOOKS
  99 #include <sys/dtrace_bsd.h>
 100 #endif
 101
 102 extern void __noinline trap(struct trapframe *frame);
 103 extern void trap_check(struct trapframe *frame);
 104 extern void syscall(struct trapframe *frame);
 105 void dblfault_handler(struct trapframe *frame);
 106
 107 static int trap_pfault(struct trapframe *, int);
 108 static void trap_fatal(struct trapframe *, vm_offset_t);
 109
 110 #define MAX_TRAP_MSG            32
 111 static char *trap_msg[] = {
 112         "",                                     /*  0 unused */
 113         "privileged instruction fault",         /*  1 T_PRIVINFLT */
 114         "",                                     /*  2 unused */
 115         "breakpoint instruction fault",         /*  3 T_BPTFLT */
 116         "",                                     /*  4 unused */
 117         "",                                     /*  5 unused */
 118         "arithmetic trap",                      /*  6 T_ARITHTRAP */
 119         "",                                     /*  7 unused */
 120         "",                                     /*  8 unused */
 121         "general protection fault",             /*  9 T_PROTFLT */
 122         "trace trap",                           /* 10 T_TRCTRAP */
 123         "",                                     /* 11 unused */
 124         "page fault",                           /* 12 T_PAGEFLT */
 125         "",                                     /* 13 unused */
 126         "alignment fault",                      /* 14 T_ALIGNFLT */
 127         "",                                     /* 15 unused */
 128         "",                                     /* 16 unused */
 129         "",                                     /* 17 unused */
 130         "integer divide fault",                 /* 18 T_DIVIDE */
 131         "non-maskable interrupt trap",          /* 19 T_NMI */
 132         "overflow trap",                        /* 20 T_OFLOW */
 133         "FPU bounds check fault",               /* 21 T_BOUND */
 134         "FPU device not available",             /* 22 T_DNA */
 135         "double fault",                         /* 23 T_DOUBLEFLT */
 136         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
 137         "invalid TSS fault",                    /* 25 T_TSSFLT */
 138         "segment not present fault",            /* 26 T_SEGNPFLT */
 139         "stack fault",                          /* 27 T_STKFLT */
 140         "machine check trap",                   /* 28 T_MCHK */
 141         "SIMD floating-point exception",        /* 29 T_XMMFLT */
 142         "reserved (unknown) fault",             /* 30 T_RESERVED */
 143         "",                                     /* 31 unused (reserved) */
 144         "DTrace pid return trap",               /* 32 T_DTRACE_RET */
 145 };
 146
 147 #ifdef KDB
 148 static int kdb_on_nmi = 1;
 149 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RWTUN,
 150         &kdb_on_nmi, 0, "Go to KDB on NMI");
 151 #endif
 152 static int panic_on_nmi = 1;
 153 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RWTUN,
 154         &panic_on_nmi, 0, "Panic on NMI");
 155 static int prot_fault_translation;
 156 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RWTUN,
 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_RWTUN,
 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         register_t dr6;
 180         int i = 0, ucode = 0, code;
 181         u_int type;
 182         register_t addr = 0;
 183         ksiginfo_t ksi;
 184
 185         PCPU_INC(cnt.v_trap);
 186         type = frame->tf_trapno;
 187
 188 #ifdef SMP
 189         /* Handler for NMI IPIs used for stopping CPUs. */
 190         if (type == T_NMI) {
 191                  if (ipi_nmi_handler() == 0)
 192                            goto out;
 193         }
 194 #endif /* SMP */
 195
 196 #ifdef KDB
 197         if (kdb_active) {
 198                 kdb_reenter();
 199                 goto out;
 200         }
 201 #endif
 202
 203         if (type == T_RESERVED) {
 204                 trap_fatal(frame, 0);
 205                 goto out;
 206         }
 207
 208         if (type == T_NMI) {
 209 #ifdef HWPMC_HOOKS
 210                 /*
 211                  * CPU PMCs interrupt using an NMI.  If the PMC module is
 212                  * active, pass the 'rip' value to the PMC module's interrupt
 213                  * handler.  A non-zero return value from the handler means that
 214                  * the NMI was consumed by it and we can return immediately.
 215                  */
 216                 if (pmc_intr != NULL &&
 217                     (*pmc_intr)(PCPU_GET(cpuid), frame) != 0)
 218                         goto out;
 219 #endif
 220
 221 #ifdef STACK
 222                 if (stack_nmi_handler(frame) != 0)
 223                         goto out;
 224 #endif
 225         }
 226
 227         if (type == T_MCHK) {
 228                 mca_intr();
 229                 goto out;
 230         }
 231
 232         if ((frame->tf_rflags & PSL_I) == 0) {
 233                 /*
 234                  * Buggy application or kernel code has disabled
 235                  * interrupts and then trapped.  Enabling interrupts
 236                  * now is wrong, but it is better than running with
 237                  * interrupts disabled until they are accidentally
 238                  * enabled later.
 239                  */
 240                 if (TRAPF_USERMODE(frame))
 241                         uprintf(
 242                             "pid %ld (%s): trap %d with interrupts disabled\n",
 243                             (long)curproc->p_pid, curthread->td_name, type);
 244                 else if (type != T_NMI && type != T_BPTFLT &&
 245                     type != T_TRCTRAP) {
 246                         /*
 247                          * XXX not quite right, since this may be for a
 248                          * multiple fault in user mode.
 249                          */
 250                         printf("kernel trap %d with interrupts disabled\n",
 251                             type);
 252
 253                         /*
 254                          * We shouldn't enable interrupts while holding a
 255                          * spin lock.
 256                          */
 257                         if (td->td_md.md_spinlock_count == 0)
 258                                 enable_intr();
 259                 }
 260         }
 261
 262         code = frame->tf_err;
 263
 264         if (TRAPF_USERMODE(frame)) {
 265                 /* user trap */
 266
 267                 td->td_pticks = 0;
 268                 td->td_frame = frame;
 269                 addr = frame->tf_rip;
 270                 if (td->td_cowgen != p->p_cowgen)
 271                         thread_cow_update(td);
 272
 273                 switch (type) {
 274                 case T_PRIVINFLT:       /* privileged instruction fault */
 275                         i = SIGILL;
 276                         ucode = ILL_PRVOPC;
 277                         break;
 278
 279                 case T_BPTFLT:          /* bpt instruction fault */
 280                 case T_TRCTRAP:         /* trace trap */
 281                         enable_intr();
 282 #ifdef KDTRACE_HOOKS
 283                         if (type == T_BPTFLT) {
 284                                 fill_frame_regs(frame, &regs);
 285                                 if (dtrace_pid_probe_ptr != NULL &&
 286                                     dtrace_pid_probe_ptr(&regs) == 0)
 287                                         goto out;
 288                         }
 289 #endif
 290                         frame->tf_rflags &= ~PSL_T;
 291                         i = SIGTRAP;
 292                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 293                         break;
 294
 295                 case T_ARITHTRAP:       /* arithmetic trap */
 296                         ucode = fputrap_x87();
 297                         if (ucode == -1)
 298                                 goto userout;
 299                         i = SIGFPE;
 300                         break;
 301
 302                 case T_PROTFLT:         /* general protection fault */
 303                         i = SIGBUS;
 304                         ucode = BUS_OBJERR;
 305                         break;
 306                 case T_STKFLT:          /* stack fault */
 307                 case T_SEGNPFLT:        /* segment not present fault */
 308                         i = SIGBUS;
 309                         ucode = BUS_ADRERR;
 310                         break;
 311                 case T_TSSFLT:          /* invalid TSS fault */
 312                         i = SIGBUS;
 313                         ucode = BUS_OBJERR;
 314                         break;
 315                 case T_ALIGNFLT:
 316                         i = SIGBUS;
 317                         ucode = BUS_ADRALN;
 318                         break;
 319                 case T_DOUBLEFLT:       /* double fault */
 320                 default:
 321                         i = SIGBUS;
 322                         ucode = BUS_OBJERR;
 323                         break;
 324
 325                 case T_PAGEFLT:         /* page fault */
 326                         /*
 327                          * Emulator can take care about this trap?
 328                          */
 329                         if (*p->p_sysent->sv_trap != NULL &&
 330                             (*p->p_sysent->sv_trap)(td) == 0)
 331                                 goto userout;
 332
 333                         addr = frame->tf_addr;
 334                         i = trap_pfault(frame, TRUE);
 335                         if (i == -1)
 336                                 goto userout;
 337                         if (i == 0)
 338                                 goto user;
 339
 340                         if (i == SIGSEGV)
 341                                 ucode = SEGV_MAPERR;
 342                         else {
 343                                 if (prot_fault_translation == 0) {
 344                                         /*
 345                                          * Autodetect.
 346                                          * This check also covers the images
 347                                          * without the ABI-tag ELF note.
 348                                          */
 349                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 350                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 351                                                 i = SIGSEGV;
 352                                                 ucode = SEGV_ACCERR;
 353                                         } else {
 354                                                 i = SIGBUS;
 355                                                 ucode = BUS_PAGE_FAULT;
 356                                         }
 357                                 } else if (prot_fault_translation == 1) {
 358                                         /*
 359                                          * Always compat mode.
 360                                          */
 361                                         i = SIGBUS;
 362                                         ucode = BUS_PAGE_FAULT;
 363                                 } else {
 364                                         /*
 365                                          * Always SIGSEGV mode.
 366                                          */
 367                                         i = SIGSEGV;
 368                                         ucode = SEGV_ACCERR;
 369                                 }
 370                         }
 371                         break;
 372
 373                 case T_DIVIDE:          /* integer divide fault */
 374                         ucode = FPE_INTDIV;
 375                         i = SIGFPE;
 376                         break;
 377
 378 #ifdef DEV_ISA
 379                 case T_NMI:
 380                         /* machine/parity/power fail/"kitchen sink" faults */
 381                         if (isa_nmi(code) == 0) {
 382 #ifdef KDB
 383                                 /*
 384                                  * NMI can be hooked up to a pushbutton
 385                                  * for debugging.
 386                                  */
 387                                 if (kdb_on_nmi) {
 388                                         printf ("NMI ... going to debugger\n");
 389                                         kdb_trap(type, 0, frame);
 390                                 }
 391 #endif /* KDB */
 392                                 goto userout;
 393                         } else if (panic_on_nmi)
 394                                 panic("NMI indicates hardware failure");
 395                         break;
 396 #endif /* DEV_ISA */
 397
 398                 case T_OFLOW:           /* integer overflow fault */
 399                         ucode = FPE_INTOVF;
 400                         i = SIGFPE;
 401                         break;
 402
 403                 case T_BOUND:           /* bounds check fault */
 404                         ucode = FPE_FLTSUB;
 405                         i = SIGFPE;
 406                         break;
 407
 408                 case T_DNA:
 409                         /* transparent fault (due to context switch "late") */
 410                         KASSERT(PCB_USER_FPU(td->td_pcb),
 411                             ("kernel FPU ctx has leaked"));
 412                         fpudna();
 413                         goto userout;
 414
 415                 case T_FPOPFLT:         /* FPU operand fetch fault */
 416                         ucode = ILL_COPROC;
 417                         i = SIGILL;
 418                         break;
 419
 420                 case T_XMMFLT:          /* SIMD floating-point exception */
 421                         ucode = fputrap_sse();
 422                         if (ucode == -1)
 423                                 goto userout;
 424                         i = SIGFPE;
 425                         break;
 426 #ifdef KDTRACE_HOOKS
 427                 case T_DTRACE_RET:
 428                         enable_intr();
 429                         fill_frame_regs(frame, &regs);
 430                         if (dtrace_return_probe_ptr != NULL &&
 431                             dtrace_return_probe_ptr(&regs) == 0)
 432                                 goto out;
 433                         break;
 434 #endif
 435                 }
 436         } else {
 437                 /* kernel trap */
 438
 439                 KASSERT(cold || td->td_ucred != NULL,
 440                     ("kernel trap doesn't have ucred"));
 441                 switch (type) {
 442                 case T_PAGEFLT:                 /* page fault */
 443                         (void) trap_pfault(frame, FALSE);
 444                         goto out;
 445
 446                 case T_DNA:
 447                         if (PCB_USER_FPU(td->td_pcb))
 448                                 panic("Unregistered use of FPU in kernel");
 449                         fpudna();
 450                         goto out;
 451
 452                 case T_ARITHTRAP:       /* arithmetic trap */
 453                 case T_XMMFLT:          /* SIMD floating-point exception */
 454                 case T_FPOPFLT:         /* FPU operand fetch fault */
 455                         /*
 456                          * For now, supporting kernel handler
 457                          * registration for FPU traps is overkill.
 458                          */
 459                         trap_fatal(frame, 0);
 460                         goto out;
 461
 462                 case T_STKFLT:          /* stack fault */
 463                 case T_PROTFLT:         /* general protection fault */
 464                 case T_SEGNPFLT:        /* segment not present fault */
 465                         if (td->td_intr_nesting_level != 0)
 466                                 break;
 467
 468                         /*
 469                          * Invalid segment selectors and out of bounds
 470                          * %rip's and %rsp's can be set up in user mode.
 471                          * This causes a fault in kernel mode when the
 472                          * kernel tries to return to user mode.  We want
 473                          * to get this fault so that we can fix the
 474                          * problem here and not have to check all the
 475                          * selectors and pointers when the user changes
 476                          * them.
 477                          */
 478                         if (frame->tf_rip == (long)doreti_iret) {
 479                                 frame->tf_rip = (long)doreti_iret_fault;
 480                                 goto out;
 481                         }
 482                         if (frame->tf_rip == (long)ld_ds) {
 483                                 frame->tf_rip = (long)ds_load_fault;
 484                                 goto out;
 485                         }
 486                         if (frame->tf_rip == (long)ld_es) {
 487                                 frame->tf_rip = (long)es_load_fault;
 488                                 goto out;
 489                         }
 490                         if (frame->tf_rip == (long)ld_fs) {
 491                                 frame->tf_rip = (long)fs_load_fault;
 492                                 goto out;
 493                         }
 494                         if (frame->tf_rip == (long)ld_gs) {
 495                                 frame->tf_rip = (long)gs_load_fault;
 496                                 goto out;
 497                         }
 498                         if (frame->tf_rip == (long)ld_gsbase) {
 499                                 frame->tf_rip = (long)gsbase_load_fault;
 500                                 goto out;
 501                         }
 502                         if (frame->tf_rip == (long)ld_fsbase) {
 503                                 frame->tf_rip = (long)fsbase_load_fault;
 504                                 goto out;
 505                         }
 506                         if (curpcb->pcb_onfault != NULL) {
 507                                 frame->tf_rip = (long)curpcb->pcb_onfault;
 508                                 goto out;
 509                         }
 510                         break;
 511
 512                 case T_TSSFLT:
 513                         /*
 514                          * PSL_NT can be set in user mode and isn't cleared
 515                          * automatically when the kernel is entered.  This
 516                          * causes a TSS fault when the kernel attempts to
 517                          * `iret' because the TSS link is uninitialized.  We
 518                          * want to get this fault so that we can fix the
 519                          * problem here and not every time the kernel is
 520                          * entered.
 521                          */
 522                         if (frame->tf_rflags & PSL_NT) {
 523                                 frame->tf_rflags &= ~PSL_NT;
 524                                 goto out;
 525                         }
 526                         break;
 527
 528                 case T_TRCTRAP:  /* trace trap */
 529                         /*
 530                          * Ignore debug register trace traps due to
 531                          * accesses in the user's address space, which
 532                          * can happen under several conditions such as
 533                          * if a user sets a watchpoint on a buffer and
 534                          * then passes that buffer to a system call.
 535                          * We still want to get TRCTRAPS for addresses
 536                          * in kernel space because that is useful when
 537                          * debugging the kernel.
 538                          */
 539                         if (user_dbreg_trap()) {
 540                                 /*
 541                                  * Reset breakpoint bits because the
 542                                  * processor doesn't
 543                                  */
 544                                 load_dr6(rdr6() & ~0xf);
 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                         /* XXX %dr6 is not quite reentrant. */
 557                         dr6 = rdr6();
 558                         load_dr6(dr6 & ~0x4000);
 559                         if (kdb_trap(type, dr6, frame))
 560                                 goto out;
 561 #endif
 562                         break;
 563
 564 #ifdef DEV_ISA
 565                 case T_NMI:
 566                         /* machine/parity/power fail/"kitchen sink" faults */
 567                         if (isa_nmi(code) == 0) {
 568 #ifdef KDB
 569                                 /*
 570                                  * NMI can be hooked up to a pushbutton
 571                                  * for debugging.
 572                                  */
 573                                 if (kdb_on_nmi) {
 574                                         printf ("NMI ... going to debugger\n");
 575                                         kdb_trap(type, 0, frame);
 576                                 }
 577 #endif /* KDB */
 578                                 goto out;
 579                         } else if (panic_on_nmi == 0)
 580                                 goto out;
 581                         /* FALLTHROUGH */
 582 #endif /* DEV_ISA */
 583                 }
 584
 585                 trap_fatal(frame, 0);
 586                 goto out;
 587         }
 588
 589         /* Translate fault for emulators (e.g. Linux) */
 590         if (*p->p_sysent->sv_transtrap)
 591                 i = (*p->p_sysent->sv_transtrap)(i, type);
 592
 593         ksiginfo_init_trap(&ksi);
 594         ksi.ksi_signo = i;
 595         ksi.ksi_code = ucode;
 596         ksi.ksi_trapno = type;
 597         ksi.ksi_addr = (void *)addr;
 598         if (uprintf_signal) {
 599                 uprintf("pid %d comm %s: signal %d err %lx code %d type %d "
 600                     "addr 0x%lx rsp 0x%lx rip 0x%lx "
 601                     "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
 602                     p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr,
 603                     frame->tf_rsp, frame->tf_rip,
 604                     fubyte((void *)(frame->tf_rip + 0)),
 605                     fubyte((void *)(frame->tf_rip + 1)),
 606                     fubyte((void *)(frame->tf_rip + 2)),
 607                     fubyte((void *)(frame->tf_rip + 3)),
 608                     fubyte((void *)(frame->tf_rip + 4)),
 609                     fubyte((void *)(frame->tf_rip + 5)),
 610                     fubyte((void *)(frame->tf_rip + 6)),
 611                     fubyte((void *)(frame->tf_rip + 7)));
 612         }
 613         KASSERT((read_rflags() & PSL_I) != 0, ("interrupts disabled"));
 614         trapsignal(td, &ksi);
 615
 616 user:
 617         userret(td, frame);
 618         KASSERT(PCB_USER_FPU(td->td_pcb),
 619             ("Return from trap with kernel FPU ctx leaked"));
 620 userout:
 621 out:
 622         return;
 623 }
 624
 625 /*
 626  * Ensure that we ignore any DTrace-induced faults. This function cannot
 627  * be instrumented, so it cannot generate such faults itself.
 628  */
 629 void
 630 trap_check(struct trapframe *frame)
 631 {
 632
 633 #ifdef KDTRACE_HOOKS
 634         if (dtrace_trap_func != NULL &&
 635             (*dtrace_trap_func)(frame, frame->tf_trapno) != 0)
 636                 return;
 637 #endif
 638         trap(frame);
 639 }
 640
 641 static int
 642 trap_pfault(frame, usermode)
 643         struct trapframe *frame;
 644         int usermode;
 645 {
 646         vm_offset_t va;
 647         vm_map_t map;
 648         int rv = 0;
 649         vm_prot_t ftype;
 650         struct thread *td = curthread;
 651         struct proc *p = td->td_proc;
 652         vm_offset_t eva = frame->tf_addr;
 653
 654         if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
 655                 /*
 656                  * Due to both processor errata and lazy TLB invalidation when
 657                  * access restrictions are removed from virtual pages, memory
 658                  * accesses that are allowed by the physical mapping layer may
 659                  * nonetheless cause one spurious page fault per virtual page.
 660                  * When the thread is executing a "no faulting" section that
 661                  * is bracketed by vm_fault_{disable,enable}_pagefaults(),
 662                  * every page fault is treated as a spurious page fault,
 663                  * unless it accesses the same virtual address as the most
 664                  * recent page fault within the same "no faulting" section.
 665                  */
 666                 if (td->td_md.md_spurflt_addr != eva ||
 667                     (td->td_pflags & TDP_RESETSPUR) != 0) {
 668                         /*
 669                          * Do nothing to the TLB.  A stale TLB entry is
 670                          * flushed automatically by a page fault.
 671                          */
 672                         td->td_md.md_spurflt_addr = eva;
 673                         td->td_pflags &= ~TDP_RESETSPUR;
 674                         return (0);
 675                 }
 676         } else {
 677                 /*
 678                  * If we get a page fault while in a critical section, then
 679                  * it is most likely a fatal kernel page fault.  The kernel
 680                  * is already going to panic trying to get a sleep lock to
 681                  * do the VM lookup, so just consider it a fatal trap so the
 682                  * kernel can print out a useful trap message and even get
 683                  * to the debugger.
 684                  *
 685                  * If we get a page fault while holding a non-sleepable
 686                  * lock, then it is most likely a fatal kernel page fault.
 687                  * If WITNESS is enabled, then it's going to whine about
 688                  * bogus LORs with various VM locks, so just skip to the
 689                  * fatal trap handling directly.
 690                  */
 691                 if (td->td_critnest != 0 ||
 692                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 693                     "Kernel page fault") != 0) {
 694                         trap_fatal(frame, eva);
 695                         return (-1);
 696                 }
 697         }
 698         va = trunc_page(eva);
 699         if (va >= VM_MIN_KERNEL_ADDRESS) {
 700                 /*
 701                  * Don't allow user-mode faults in kernel address space.
 702                  */
 703                 if (usermode)
 704                         goto nogo;
 705
 706                 map = kernel_map;
 707         } else {
 708                 map = &p->p_vmspace->vm_map;
 709
 710                 /*
 711                  * When accessing a usermode address, kernel must be
 712                  * ready to accept the page fault, and provide a
 713                  * handling routine.  Since accessing the address
 714                  * without the handler is a bug, do not try to handle
 715                  * it normally, and panic immediately.
 716                  */
 717                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 718                     curpcb->pcb_onfault == NULL)) {
 719                         trap_fatal(frame, eva);
 720                         return (-1);
 721                 }
 722         }
 723
 724         /*
 725          * If the trap was caused by errant bits in the PTE then panic.
 726          */
 727         if (frame->tf_err & PGEX_RSV) {
 728                 trap_fatal(frame, eva);
 729                 return (-1);
 730         }
 731
 732         /*
 733          * PGEX_I is defined only if the execute disable bit capability is
 734          * supported and enabled.
 735          */
 736         if (frame->tf_err & PGEX_W)
 737                 ftype = VM_PROT_WRITE;
 738         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 739                 ftype = VM_PROT_EXECUTE;
 740         else
 741                 ftype = VM_PROT_READ;
 742
 743         /* Fault in the page. */
 744         rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 745         if (rv == KERN_SUCCESS) {
 746 #ifdef HWPMC_HOOKS
 747                 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
 748                         PMC_SOFT_CALL_TF( , , page_fault, all, frame);
 749                         if (ftype == VM_PROT_READ)
 750                                 PMC_SOFT_CALL_TF( , , page_fault, read,
 751                                     frame);
 752                         else
 753                                 PMC_SOFT_CALL_TF( , , page_fault, write,
 754                                     frame);
 755                 }
 756 #endif
 757                 return (0);
 758         }
 759 nogo:
 760         if (!usermode) {
 761                 if (td->td_intr_nesting_level == 0 &&
 762                     curpcb->pcb_onfault != NULL) {
 763                         frame->tf_rip = (long)curpcb->pcb_onfault;
 764                         return (0);
 765                 }
 766                 trap_fatal(frame, eva);
 767                 return (-1);
 768         }
 769         return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 770 }
 771
 772 static void
 773 trap_fatal(frame, eva)
 774         struct trapframe *frame;
 775         vm_offset_t eva;
 776 {
 777         int code, ss;
 778         u_int type;
 779         long esp;
 780         struct soft_segment_descriptor softseg;
 781         char *msg;
 782
 783         code = frame->tf_err;
 784         type = frame->tf_trapno;
 785         sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)],
 786             &softseg);
 787
 788         if (type <= MAX_TRAP_MSG)
 789                 msg = trap_msg[type];
 790         else
 791                 msg = "UNKNOWN";
 792         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 793             TRAPF_USERMODE(frame) ? "user" : "kernel");
 794 #ifdef SMP
 795         /* two separate prints in case of a trap on an unmapped page */
 796         printf("cpuid = %d; ", PCPU_GET(cpuid));
 797         printf("apic id = %02x\n", PCPU_GET(apic_id));
 798 #endif
 799         if (type == T_PAGEFLT) {
 800                 printf("fault virtual address   = 0x%lx\n", eva);
 801                 printf("fault code              = %s %s %s%s, %s\n",
 802                         code & PGEX_U ? "user" : "supervisor",
 803                         code & PGEX_W ? "write" : "read",
 804                         code & PGEX_I ? "instruction" : "data",
 805                         code & PGEX_RSV ? " rsv" : "",
 806                         code & PGEX_P ? "protection violation" : "page not present");
 807         }
 808         printf("instruction pointer     = 0x%lx:0x%lx\n",
 809                frame->tf_cs & 0xffff, frame->tf_rip);
 810         if (TF_HAS_STACKREGS(frame)) {
 811                 ss = frame->tf_ss & 0xffff;
 812                 esp = frame->tf_rsp;
 813         } else {
 814                 ss = GSEL(GDATA_SEL, SEL_KPL);
 815                 esp = (long)&frame->tf_rsp;
 816         }
 817         printf("stack pointer           = 0x%x:0x%lx\n", ss, esp);
 818         printf("frame pointer           = 0x%x:0x%lx\n", ss, frame->tf_rbp);
 819         printf("code segment            = base 0x%lx, limit 0x%lx, type 0x%x\n",
 820                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 821         printf("                        = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
 822                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
 823                softseg.ssd_gran);
 824         printf("processor eflags        = ");
 825         if (frame->tf_rflags & PSL_T)
 826                 printf("trace trap, ");
 827         if (frame->tf_rflags & PSL_I)
 828                 printf("interrupt enabled, ");
 829         if (frame->tf_rflags & PSL_NT)
 830                 printf("nested task, ");
 831         if (frame->tf_rflags & PSL_RF)
 832                 printf("resume, ");
 833         printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
 834         printf("current process         = %d (%s)\n",
 835             curproc->p_pid, curthread->td_name);
 836
 837 #ifdef KDB
 838         if (debugger_on_panic || kdb_active)
 839                 if (kdb_trap(type, 0, frame))
 840                         return;
 841 #endif
 842         printf("trap number             = %d\n", type);
 843         if (type <= MAX_TRAP_MSG)
 844                 panic("%s", trap_msg[type]);
 845         else
 846                 panic("unknown/reserved trap");
 847 }
 848
 849 /*
 850  * Double fault handler. Called when a fault occurs while writing
 851  * a frame for a trap/exception onto the stack. This usually occurs
 852  * when the stack overflows (such is the case with infinite recursion,
 853  * for example).
 854  */
 855 void
 856 dblfault_handler(struct trapframe *frame)
 857 {
 858 #ifdef KDTRACE_HOOKS
 859         if (dtrace_doubletrap_func != NULL)
 860                 (*dtrace_doubletrap_func)();
 861 #endif
 862         printf("\nFatal double fault\n");
 863         printf("rip = 0x%lx\n", frame->tf_rip);
 864         printf("rsp = 0x%lx\n", frame->tf_rsp);
 865         printf("rbp = 0x%lx\n", frame->tf_rbp);
 866 #ifdef SMP
 867         /* two separate prints in case of a trap on an unmapped page */
 868         printf("cpuid = %d; ", PCPU_GET(cpuid));
 869         printf("apic id = %02x\n", PCPU_GET(apic_id));
 870 #endif
 871         panic("double fault");
 872 }
 873
 874 int
 875 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
 876 {
 877         struct proc *p;
 878         struct trapframe *frame;
 879         register_t *argp;
 880         caddr_t params;
 881         int reg, regcnt, error;
 882
 883         p = td->td_proc;
 884         frame = td->td_frame;
 885         reg = 0;
 886         regcnt = 6;
 887
 888         params = (caddr_t)frame->tf_rsp + sizeof(register_t);
 889         sa->code = frame->tf_rax;
 890
 891         if (sa->code == SYS_syscall || sa->code == SYS___syscall) {
 892                 sa->code = frame->tf_rdi;
 893                 reg++;
 894                 regcnt--;
 895         }
 896         if (p->p_sysent->sv_mask)
 897                 sa->code &= p->p_sysent->sv_mask;
 898
 899         if (sa->code >= p->p_sysent->sv_size)
 900                 sa->callp = &p->p_sysent->sv_table[0];
 901         else
 902                 sa->callp = &p->p_sysent->sv_table[sa->code];
 903
 904         sa->narg = sa->callp->sy_narg;
 905         KASSERT(sa->narg <= sizeof(sa->args) / sizeof(sa->args[0]),
 906             ("Too many syscall arguments!"));
 907         error = 0;
 908         argp = &frame->tf_rdi;
 909         argp += reg;
 910         bcopy(argp, sa->args, sizeof(sa->args[0]) * regcnt);
 911         if (sa->narg > regcnt) {
 912                 KASSERT(params != NULL, ("copyin args with no params!"));
 913                 error = copyin(params, &sa->args[regcnt],
 914                     (sa->narg - regcnt) * sizeof(sa->args[0]));
 915         }
 916
 917         if (error == 0) {
 918                 td->td_retval[0] = 0;
 919                 td->td_retval[1] = frame->tf_rdx;
 920         }
 921
 922         return (error);
 923 }
 924
 925 #include "../../kern/subr_syscall.c"
 926
 927 /*
 928  * System call handler for native binaries.  The trap frame is already
 929  * set up by the assembler trampoline and a pointer to it is saved in
 930  * td_frame.
 931  */
 932 void
 933 amd64_syscall(struct thread *td, int traced)
 934 {
 935         struct syscall_args sa;
 936         int error;
 937         ksiginfo_t ksi;
 938
 939 #ifdef DIAGNOSTIC
 940         if (!TRAPF_USERMODE(td->td_frame)) {
 941                 panic("syscall");
 942                 /* NOT REACHED */
 943         }
 944 #endif
 945         error = syscallenter(td, &sa);
 946
 947         /*
 948          * Traced syscall.
 949          */
 950         if (__predict_false(traced)) {
 951                 td->td_frame->tf_rflags &= ~PSL_T;
 952                 ksiginfo_init_trap(&ksi);
 953                 ksi.ksi_signo = SIGTRAP;
 954                 ksi.ksi_code = TRAP_TRACE;
 955                 ksi.ksi_addr = (void *)td->td_frame->tf_rip;
 956                 trapsignal(td, &ksi);
 957         }
 958
 959         KASSERT(PCB_USER_FPU(td->td_pcb),
 960             ("System call %s returing with kernel FPU ctx leaked",
 961              syscallname(td->td_proc, sa.code)));
 962         KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td),
 963             ("System call %s returning with mangled pcb_save",
 964              syscallname(td->td_proc, sa.code)));
 965         KASSERT(td->td_md.md_invl_gen.gen == 0,
 966             ("System call %s returning with leaked invl_gen %lu",
 967             syscallname(td->td_proc, sa.code), td->td_md.md_invl_gen.gen));
 968
 969
 970         syscallret(td, error, &sa);
 971
 972         /*
 973          * If the user-supplied value of %rip is not a canonical
 974          * address, then some CPUs will trigger a ring 0 #GP during
 975          * the sysret instruction.  However, the fault handler would
 976          * execute in ring 0 with the user's %gs and %rsp which would
 977          * not be safe.  Instead, use the full return path which
 978          * catches the problem safely.
 979          */
 980         if (td->td_frame->tf_rip >= VM_MAXUSER_ADDRESS)
 981                 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
 982 }