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