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                         /*
 326                          * Emulator can take care about this trap?
 327                          */
 328                         if (*p->p_sysent->sv_trap != NULL &&
 329                             (*p->p_sysent->sv_trap)(td) == 0)
 330                                 goto userout;
 331
 332                         addr = frame->tf_addr;
 333                         i = trap_pfault(frame, TRUE);
 334                         if (i == -1)
 335                                 goto userout;
 336                         if (i == 0)
 337                                 goto user;
 338
 339                         if (i == SIGSEGV)
 340                                 ucode = SEGV_MAPERR;
 341                         else {
 342                                 if (prot_fault_translation == 0) {
 343                                         /*
 344                                          * Autodetect.
 345                                          * This check also covers the images
 346                                          * without the ABI-tag ELF note.
 347                                          */
 348                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 349                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 350                                                 i = SIGSEGV;
 351                                                 ucode = SEGV_ACCERR;
 352                                         } else {
 353                                                 i = SIGBUS;
 354                                                 ucode = BUS_PAGE_FAULT;
 355                                         }
 356                                 } else if (prot_fault_translation == 1) {
 357                                         /*
 358                                          * Always compat mode.
 359                                          */
 360                                         i = SIGBUS;
 361                                         ucode = BUS_PAGE_FAULT;
 362                                 } else {
 363                                         /*
 364                                          * Always SIGSEGV mode.
 365                                          */
 366                                         i = SIGSEGV;
 367                                         ucode = SEGV_ACCERR;
 368                                 }
 369                         }
 370                         break;
 371
 372                 case T_DIVIDE:          /* integer divide fault */
 373                         ucode = FPE_INTDIV;
 374                         i = SIGFPE;
 375                         break;
 376
 377 #ifdef DEV_ISA
 378                 case T_NMI:
 379                         /* machine/parity/power fail/"kitchen sink" faults */
 380                         if (isa_nmi(code) == 0) {
 381 #ifdef KDB
 382                                 /*
 383                                  * NMI can be hooked up to a pushbutton
 384                                  * for debugging.
 385                                  */
 386                                 if (kdb_on_nmi) {
 387                                         printf ("NMI ... going to debugger\n");
 388                                         kdb_trap(type, 0, frame);
 389                                 }
 390 #endif /* KDB */
 391                                 goto userout;
 392                         } else if (panic_on_nmi)
 393                                 panic("NMI indicates hardware failure");
 394                         break;
 395 #endif /* DEV_ISA */
 396
 397                 case T_OFLOW:           /* integer overflow fault */
 398                         ucode = FPE_INTOVF;
 399                         i = SIGFPE;
 400                         break;
 401
 402                 case T_BOUND:           /* bounds check fault */
 403                         ucode = FPE_FLTSUB;
 404                         i = SIGFPE;
 405                         break;
 406
 407                 case T_DNA:
 408                         /* transparent fault (due to context switch "late") */
 409                         KASSERT(PCB_USER_FPU(td->td_pcb),
 410                             ("kernel FPU ctx has leaked"));
 411                         fpudna();
 412                         goto userout;
 413
 414                 case T_FPOPFLT:         /* FPU operand fetch fault */
 415                         ucode = ILL_COPROC;
 416                         i = SIGILL;
 417                         break;
 418
 419                 case T_XMMFLT:          /* SIMD floating-point exception */
 420                         ucode = fputrap_sse();
 421                         if (ucode == -1)
 422                                 goto userout;
 423                         i = SIGFPE;
 424                         break;
 425 #ifdef KDTRACE_HOOKS
 426                 case T_DTRACE_RET:
 427                         enable_intr();
 428                         fill_frame_regs(frame, &regs);
 429                         if (dtrace_return_probe_ptr != NULL &&
 430                             dtrace_return_probe_ptr(&regs) == 0)
 431                                 goto out;
 432                         break;
 433 #endif
 434                 }
 435         } else {
 436                 /* kernel trap */
 437
 438                 KASSERT(cold || td->td_ucred != NULL,
 439                     ("kernel trap doesn't have ucred"));
 440                 switch (type) {
 441                 case T_PAGEFLT:                 /* page fault */
 442                         (void) trap_pfault(frame, FALSE);
 443                         goto out;
 444
 445                 case T_DNA:
 446                         KASSERT(!PCB_USER_FPU(td->td_pcb),
 447                             ("Unregistered use of FPU in kernel"));
 448                         fpudna();
 449                         goto out;
 450
 451                 case T_ARITHTRAP:       /* arithmetic trap */
 452                 case T_XMMFLT:          /* SIMD floating-point exception */
 453                 case T_FPOPFLT:         /* FPU operand fetch fault */
 454                         /*
 455                          * For now, supporting kernel handler
 456                          * registration for FPU traps is overkill.
 457                          */
 458                         trap_fatal(frame, 0);
 459                         goto out;
 460
 461                 case T_STKFLT:          /* stack fault */
 462                 case T_PROTFLT:         /* general protection fault */
 463                 case T_SEGNPFLT:        /* segment not present fault */
 464                         if (td->td_intr_nesting_level != 0)
 465                                 break;
 466
 467                         /*
 468                          * Invalid segment selectors and out of bounds
 469                          * %rip's and %rsp's can be set up in user mode.
 470                          * This causes a fault in kernel mode when the
 471                          * kernel tries to return to user mode.  We want
 472                          * to get this fault so that we can fix the
 473                          * problem here and not have to check all the
 474                          * selectors and pointers when the user changes
 475                          * them.
 476                          */
 477                         if (frame->tf_rip == (long)doreti_iret) {
 478                                 frame->tf_rip = (long)doreti_iret_fault;
 479                                 goto out;
 480                         }
 481                         if (frame->tf_rip == (long)ld_ds) {
 482                                 frame->tf_rip = (long)ds_load_fault;
 483                                 goto out;
 484                         }
 485                         if (frame->tf_rip == (long)ld_es) {
 486                                 frame->tf_rip = (long)es_load_fault;
 487                                 goto out;
 488                         }
 489                         if (frame->tf_rip == (long)ld_fs) {
 490                                 frame->tf_rip = (long)fs_load_fault;
 491                                 goto out;
 492                         }
 493                         if (frame->tf_rip == (long)ld_gs) {
 494                                 frame->tf_rip = (long)gs_load_fault;
 495                                 goto out;
 496                         }
 497                         if (frame->tf_rip == (long)ld_gsbase) {
 498                                 frame->tf_rip = (long)gsbase_load_fault;
 499                                 goto out;
 500                         }
 501                         if (frame->tf_rip == (long)ld_fsbase) {
 502                                 frame->tf_rip = (long)fsbase_load_fault;
 503                                 goto out;
 504                         }
 505                         if (curpcb->pcb_onfault != NULL) {
 506                                 frame->tf_rip = (long)curpcb->pcb_onfault;
 507                                 goto out;
 508                         }
 509                         break;
 510
 511                 case T_TSSFLT:
 512                         /*
 513                          * PSL_NT can be set in user mode and isn't cleared
 514                          * automatically when the kernel is entered.  This
 515                          * causes a TSS fault when the kernel attempts to
 516                          * `iret' because the TSS link is uninitialized.  We
 517                          * want to get this fault so that we can fix the
 518                          * problem here and not every time the kernel is
 519                          * entered.
 520                          */
 521                         if (frame->tf_rflags & PSL_NT) {
 522                                 frame->tf_rflags &= ~PSL_NT;
 523                                 goto out;
 524                         }
 525                         break;
 526
 527                 case T_TRCTRAP:  /* trace trap */
 528                         /*
 529                          * Ignore debug register trace traps due to
 530                          * accesses in the user's address space, which
 531                          * can happen under several conditions such as
 532                          * if a user sets a watchpoint on a buffer and
 533                          * then passes that buffer to a system call.
 534                          * We still want to get TRCTRAPS for addresses
 535                          * in kernel space because that is useful when
 536                          * debugging the kernel.
 537                          */
 538                         if (user_dbreg_trap()) {
 539                                 /*
 540                                  * Reset breakpoint bits because the
 541                                  * processor doesn't
 542                                  */
 543                                 /* XXX check upper bits here */
 544                                 load_dr6(rdr6() & 0xfffffff0);
 545                                 goto out;
 546                         }
 547                         /*
 548                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 549                          */
 550                 case T_BPTFLT:
 551                         /*
 552                          * If KDB is enabled, let it handle the debugger trap.
 553                          * Otherwise, debugger traps "can't happen".
 554                          */
 555 #ifdef KDB
 556                         if (kdb_trap(type, 0, frame))
 557                                 goto out;
 558 #endif
 559                         break;
 560
 561 #ifdef DEV_ISA
 562                 case T_NMI:
 563                         /* machine/parity/power fail/"kitchen sink" faults */
 564                         if (isa_nmi(code) == 0) {
 565 #ifdef KDB
 566                                 /*
 567                                  * NMI can be hooked up to a pushbutton
 568                                  * for debugging.
 569                                  */
 570                                 if (kdb_on_nmi) {
 571                                         printf ("NMI ... going to debugger\n");
 572                                         kdb_trap(type, 0, frame);
 573                                 }
 574 #endif /* KDB */
 575                                 goto out;
 576                         } else if (panic_on_nmi == 0)
 577                                 goto out;
 578                         /* FALLTHROUGH */
 579 #endif /* DEV_ISA */
 580                 }
 581
 582                 trap_fatal(frame, 0);
 583                 goto out;
 584         }
 585
 586         /* Translate fault for emulators (e.g. Linux) */
 587         if (*p->p_sysent->sv_transtrap)
 588                 i = (*p->p_sysent->sv_transtrap)(i, type);
 589
 590         ksiginfo_init_trap(&ksi);
 591         ksi.ksi_signo = i;
 592         ksi.ksi_code = ucode;
 593         ksi.ksi_trapno = type;
 594         ksi.ksi_addr = (void *)addr;
 595         if (uprintf_signal) {
 596                 uprintf("pid %d comm %s: signal %d err %lx code %d type %d "
 597                     "addr 0x%lx rsp 0x%lx rip 0x%lx "
 598                     "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
 599                     p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr,
 600                     frame->tf_rsp, frame->tf_rip,
 601                     fubyte((void *)(frame->tf_rip + 0)),
 602                     fubyte((void *)(frame->tf_rip + 1)),
 603                     fubyte((void *)(frame->tf_rip + 2)),
 604                     fubyte((void *)(frame->tf_rip + 3)),
 605                     fubyte((void *)(frame->tf_rip + 4)),
 606                     fubyte((void *)(frame->tf_rip + 5)),
 607                     fubyte((void *)(frame->tf_rip + 6)),
 608                     fubyte((void *)(frame->tf_rip + 7)));
 609         }
 610         KASSERT((read_rflags() & PSL_I) != 0, ("interrupts disabled"));
 611         trapsignal(td, &ksi);
 612
 613 user:
 614         userret(td, frame);
 615         KASSERT(PCB_USER_FPU(td->td_pcb),
 616             ("Return from trap with kernel FPU ctx leaked"));
 617 userout:
 618 out:
 619         return;
 620 }
 621
 622 /*
 623  * Ensure that we ignore any DTrace-induced faults. This function cannot
 624  * be instrumented, so it cannot generate such faults itself.
 625  */
 626 void
 627 trap_check(struct trapframe *frame)
 628 {
 629
 630 #ifdef KDTRACE_HOOKS
 631         if (dtrace_trap_func != NULL &&
 632             (*dtrace_trap_func)(frame, frame->tf_trapno) != 0)
 633                 return;
 634 #endif
 635         trap(frame);
 636 }
 637
 638 static int
 639 trap_pfault(frame, usermode)
 640         struct trapframe *frame;
 641         int usermode;
 642 {
 643         vm_offset_t va;
 644         vm_map_t map;
 645         int rv = 0;
 646         vm_prot_t ftype;
 647         struct thread *td = curthread;
 648         struct proc *p = td->td_proc;
 649         vm_offset_t eva = frame->tf_addr;
 650
 651         if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
 652                 /*
 653                  * Due to both processor errata and lazy TLB invalidation when
 654                  * access restrictions are removed from virtual pages, memory
 655                  * accesses that are allowed by the physical mapping layer may
 656                  * nonetheless cause one spurious page fault per virtual page.
 657                  * When the thread is executing a "no faulting" section that
 658                  * is bracketed by vm_fault_{disable,enable}_pagefaults(),
 659                  * every page fault is treated as a spurious page fault,
 660                  * unless it accesses the same virtual address as the most
 661                  * recent page fault within the same "no faulting" section.
 662                  */
 663                 if (td->td_md.md_spurflt_addr != eva ||
 664                     (td->td_pflags & TDP_RESETSPUR) != 0) {
 665                         /*
 666                          * Do nothing to the TLB.  A stale TLB entry is
 667                          * flushed automatically by a page fault.
 668                          */
 669                         td->td_md.md_spurflt_addr = eva;
 670                         td->td_pflags &= ~TDP_RESETSPUR;
 671                         return (0);
 672                 }
 673         } else {
 674                 /*
 675                  * If we get a page fault while in a critical section, then
 676                  * it is most likely a fatal kernel page fault.  The kernel
 677                  * is already going to panic trying to get a sleep lock to
 678                  * do the VM lookup, so just consider it a fatal trap so the
 679                  * kernel can print out a useful trap message and even get
 680                  * to the debugger.
 681                  *
 682                  * If we get a page fault while holding a non-sleepable
 683                  * lock, then it is most likely a fatal kernel page fault.
 684                  * If WITNESS is enabled, then it's going to whine about
 685                  * bogus LORs with various VM locks, so just skip to the
 686                  * fatal trap handling directly.
 687                  */
 688                 if (td->td_critnest != 0 ||
 689                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 690                     "Kernel page fault") != 0) {
 691                         trap_fatal(frame, eva);
 692                         return (-1);
 693                 }
 694         }
 695         va = trunc_page(eva);
 696         if (va >= VM_MIN_KERNEL_ADDRESS) {
 697                 /*
 698                  * Don't allow user-mode faults in kernel address space.
 699                  */
 700                 if (usermode)
 701                         goto nogo;
 702
 703                 map = kernel_map;
 704         } else {
 705                 map = &p->p_vmspace->vm_map;
 706
 707                 /*
 708                  * When accessing a usermode address, kernel must be
 709                  * ready to accept the page fault, and provide a
 710                  * handling routine.  Since accessing the address
 711                  * without the handler is a bug, do not try to handle
 712                  * it normally, and panic immediately.
 713                  */
 714                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 715                     curpcb->pcb_onfault == NULL)) {
 716                         trap_fatal(frame, eva);
 717                         return (-1);
 718                 }
 719         }
 720
 721         /*
 722          * If the trap was caused by errant bits in the PTE then panic.
 723          */
 724         if (frame->tf_err & PGEX_RSV) {
 725                 trap_fatal(frame, eva);
 726                 return (-1);
 727         }
 728
 729         /*
 730          * PGEX_I is defined only if the execute disable bit capability is
 731          * supported and enabled.
 732          */
 733         if (frame->tf_err & PGEX_W)
 734                 ftype = VM_PROT_WRITE;
 735         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 736                 ftype = VM_PROT_EXECUTE;
 737         else
 738                 ftype = VM_PROT_READ;
 739
 740         /* Fault in the page. */
 741         rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 742         if (rv == KERN_SUCCESS) {
 743 #ifdef HWPMC_HOOKS
 744                 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
 745                         PMC_SOFT_CALL_TF( , , page_fault, all, frame);
 746                         if (ftype == VM_PROT_READ)
 747                                 PMC_SOFT_CALL_TF( , , page_fault, read,
 748                                     frame);
 749                         else
 750                                 PMC_SOFT_CALL_TF( , , page_fault, write,
 751                                     frame);
 752                 }
 753 #endif
 754                 return (0);
 755         }
 756 nogo:
 757         if (!usermode) {
 758                 if (td->td_intr_nesting_level == 0 &&
 759                     curpcb->pcb_onfault != NULL) {
 760                         frame->tf_rip = (long)curpcb->pcb_onfault;
 761                         return (0);
 762                 }
 763                 trap_fatal(frame, eva);
 764                 return (-1);
 765         }
 766         return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 767 }
 768
 769 static void
 770 trap_fatal(frame, eva)
 771         struct trapframe *frame;
 772         vm_offset_t eva;
 773 {
 774         int code, ss;
 775         u_int type;
 776         long esp;
 777         struct soft_segment_descriptor softseg;
 778         char *msg;
 779
 780         code = frame->tf_err;
 781         type = frame->tf_trapno;
 782         sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)],
 783             &softseg);
 784
 785         if (type <= MAX_TRAP_MSG)
 786                 msg = trap_msg[type];
 787         else
 788                 msg = "UNKNOWN";
 789         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 790             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
 791 #ifdef SMP
 792         /* two separate prints in case of a trap on an unmapped page */
 793         printf("cpuid = %d; ", PCPU_GET(cpuid));
 794         printf("apic id = %02x\n", PCPU_GET(apic_id));
 795 #endif
 796         if (type == T_PAGEFLT) {
 797                 printf("fault virtual address   = 0x%lx\n", eva);
 798                 printf("fault code              = %s %s %s%s, %s\n",
 799                         code & PGEX_U ? "user" : "supervisor",
 800                         code & PGEX_W ? "write" : "read",
 801                         code & PGEX_I ? "instruction" : "data",
 802                         code & PGEX_RSV ? " rsv" : "",
 803                         code & PGEX_P ? "protection violation" : "page not present");
 804         }
 805         printf("instruction pointer     = 0x%lx:0x%lx\n",
 806                frame->tf_cs & 0xffff, frame->tf_rip);
 807         if (ISPL(frame->tf_cs) == SEL_UPL) {
 808                 ss = frame->tf_ss & 0xffff;
 809                 esp = frame->tf_rsp;
 810         } else {
 811                 ss = GSEL(GDATA_SEL, SEL_KPL);
 812                 esp = (long)&frame->tf_rsp;
 813         }
 814         printf("stack pointer           = 0x%x:0x%lx\n", ss, esp);
 815         printf("frame pointer           = 0x%x:0x%lx\n", ss, frame->tf_rbp);
 816         printf("code segment            = base 0x%lx, limit 0x%lx, type 0x%x\n",
 817                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 818         printf("                        = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
 819                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
 820                softseg.ssd_gran);
 821         printf("processor eflags        = ");
 822         if (frame->tf_rflags & PSL_T)
 823                 printf("trace trap, ");
 824         if (frame->tf_rflags & PSL_I)
 825                 printf("interrupt enabled, ");
 826         if (frame->tf_rflags & PSL_NT)
 827                 printf("nested task, ");
 828         if (frame->tf_rflags & PSL_RF)
 829                 printf("resume, ");
 830         printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
 831         printf("current process         = %d (%s)\n",
 832             curproc->p_pid, curthread->td_name);
 833
 834 #ifdef KDB
 835         if (debugger_on_panic || kdb_active)
 836                 if (kdb_trap(type, 0, frame))
 837                         return;
 838 #endif
 839         printf("trap number             = %d\n", type);
 840         if (type <= MAX_TRAP_MSG)
 841                 panic("%s", trap_msg[type]);
 842         else
 843                 panic("unknown/reserved trap");
 844 }
 845
 846 /*
 847  * Double fault handler. Called when a fault occurs while writing
 848  * a frame for a trap/exception onto the stack. This usually occurs
 849  * when the stack overflows (such is the case with infinite recursion,
 850  * for example).
 851  */
 852 void
 853 dblfault_handler(struct trapframe *frame)
 854 {
 855 #ifdef KDTRACE_HOOKS
 856         if (dtrace_doubletrap_func != NULL)
 857                 (*dtrace_doubletrap_func)();
 858 #endif
 859         printf("\nFatal double fault\n");
 860         printf("rip = 0x%lx\n", frame->tf_rip);
 861         printf("rsp = 0x%lx\n", frame->tf_rsp);
 862         printf("rbp = 0x%lx\n", frame->tf_rbp);
 863 #ifdef SMP
 864         /* two separate prints in case of a trap on an unmapped page */
 865         printf("cpuid = %d; ", PCPU_GET(cpuid));
 866         printf("apic id = %02x\n", PCPU_GET(apic_id));
 867 #endif
 868         panic("double fault");
 869 }
 870
 871 int
 872 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
 873 {
 874         struct proc *p;
 875         struct trapframe *frame;
 876         register_t *argp;
 877         caddr_t params;
 878         int reg, regcnt, error;
 879
 880         p = td->td_proc;
 881         frame = td->td_frame;
 882         reg = 0;
 883         regcnt = 6;
 884
 885         params = (caddr_t)frame->tf_rsp + sizeof(register_t);
 886         sa->code = frame->tf_rax;
 887
 888         if (sa->code == SYS_syscall || sa->code == SYS___syscall) {
 889                 sa->code = frame->tf_rdi;
 890                 reg++;
 891                 regcnt--;
 892         }
 893         if (p->p_sysent->sv_mask)
 894                 sa->code &= p->p_sysent->sv_mask;
 895
 896         if (sa->code >= p->p_sysent->sv_size)
 897                 sa->callp = &p->p_sysent->sv_table[0];
 898         else
 899                 sa->callp = &p->p_sysent->sv_table[sa->code];
 900
 901         sa->narg = sa->callp->sy_narg;
 902         KASSERT(sa->narg <= sizeof(sa->args) / sizeof(sa->args[0]),
 903             ("Too many syscall arguments!"));
 904         error = 0;
 905         argp = &frame->tf_rdi;
 906         argp += reg;
 907         bcopy(argp, sa->args, sizeof(sa->args[0]) * regcnt);
 908         if (sa->narg > regcnt) {
 909                 KASSERT(params != NULL, ("copyin args with no params!"));
 910                 error = copyin(params, &sa->args[regcnt],
 911                     (sa->narg - regcnt) * sizeof(sa->args[0]));
 912         }
 913
 914         if (error == 0) {
 915                 td->td_retval[0] = 0;
 916                 td->td_retval[1] = frame->tf_rdx;
 917         }
 918
 919         return (error);
 920 }
 921
 922 #include "../../kern/subr_syscall.c"
 923
 924 /*
 925  * System call handler for native binaries.  The trap frame is already
 926  * set up by the assembler trampoline and a pointer to it is saved in
 927  * td_frame.
 928  */
 929 void
 930 amd64_syscall(struct thread *td, int traced)
 931 {
 932         struct syscall_args sa;
 933         int error;
 934         ksiginfo_t ksi;
 935
 936 #ifdef DIAGNOSTIC
 937         if (ISPL(td->td_frame->tf_cs) != SEL_UPL) {
 938                 panic("syscall");
 939                 /* NOT REACHED */
 940         }
 941 #endif
 942         error = syscallenter(td, &sa);
 943
 944         /*
 945          * Traced syscall.
 946          */
 947         if (__predict_false(traced)) {
 948                 td->td_frame->tf_rflags &= ~PSL_T;
 949                 ksiginfo_init_trap(&ksi);
 950                 ksi.ksi_signo = SIGTRAP;
 951                 ksi.ksi_code = TRAP_TRACE;
 952                 ksi.ksi_addr = (void *)td->td_frame->tf_rip;
 953                 trapsignal(td, &ksi);
 954         }
 955
 956         KASSERT(PCB_USER_FPU(td->td_pcb),
 957             ("System call %s returing with kernel FPU ctx leaked",
 958              syscallname(td->td_proc, sa.code)));
 959         KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td),
 960             ("System call %s returning with mangled pcb_save",
 961              syscallname(td->td_proc, sa.code)));
 962
 963         syscallret(td, error, &sa);
 964
 965         /*
 966          * If the user-supplied value of %rip is not a canonical
 967          * address, then some CPUs will trigger a ring 0 #GP during
 968          * the sysret instruction.  However, the fault handler would
 969          * execute in ring 0 with the user's %gs and %rsp which would
 970          * not be safe.  Instead, use the full return path which
 971          * catches the problem safely.
 972          */
 973         if (td->td_frame->tf_rip >= VM_MAXUSER_ADDRESS)
 974                 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
 975 }