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