sys/amd64/amd64/trap.c

   1 /*-
   2  * Copyright (C) 1994, David Greenman
   3  * Copyright (c) 1990, 1993
   4  *      The Regents of the University of California.  All rights reserved.
   5  *
   6  * This code is derived from software contributed to Berkeley by
   7  * the University of Utah, and William Jolitz.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in the
  16  *    documentation and/or other materials provided with the distribution.
  17  * 3. All advertising materials mentioning features or use of this software
  18  *    must display the following acknowledgement:
  19  *      This product includes software developed by the University of
  20  *      California, Berkeley and its contributors.
  21  * 4. Neither the name of the University nor the names of its contributors
  22  *    may be used to endorse or promote products derived from this software
  23  *    without specific prior written permission.
  24  *
  25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  35  * SUCH DAMAGE.
  36  *
  37  *      from: @(#)trap.c        7.4 (Berkeley) 5/13/91
  38  */
  39
  40 #include <sys/cdefs.h>
  41 __FBSDID("$FreeBSD$");
  42
  43 /*
  44  * AMD64 Trap and System call handling
  45  */
  46
  47 #include "opt_clock.h"
  48 #include "opt_cpu.h"
  49 #include "opt_hwpmc_hooks.h"
  50 #include "opt_isa.h"
  51 #include "opt_kdb.h"
  52 #include "opt_kdtrace.h"
  53
  54 #include <sys/param.h>
  55 #include <sys/bus.h>
  56 #include <sys/systm.h>
  57 #include <sys/proc.h>
  58 #include <sys/pioctl.h>
  59 #include <sys/ptrace.h>
  60 #include <sys/kdb.h>
  61 #include <sys/kernel.h>
  62 #include <sys/ktr.h>
  63 #include <sys/lock.h>
  64 #include <sys/mutex.h>
  65 #include <sys/resourcevar.h>
  66 #include <sys/signalvar.h>
  67 #include <sys/syscall.h>
  68 #include <sys/sysctl.h>
  69 #include <sys/sysent.h>
  70 #include <sys/uio.h>
  71 #include <sys/vmmeter.h>
  72 #ifdef HWPMC_HOOKS
  73 #include <sys/pmckern.h>
  74 PMC_SOFT_DEFINE( , , page_fault, all);
  75 PMC_SOFT_DEFINE( , , page_fault, read);
  76 PMC_SOFT_DEFINE( , , page_fault, write);
  77 #endif
  78
  79 #include <vm/vm.h>
  80 #include <vm/vm_param.h>
  81 #include <vm/pmap.h>
  82 #include <vm/vm_kern.h>
  83 #include <vm/vm_map.h>
  84 #include <vm/vm_page.h>
  85 #include <vm/vm_extern.h>
  86
  87 #include <machine/cpu.h>
  88 #include <machine/intr_machdep.h>
  89 #include <x86/mca.h>
  90 #include <machine/md_var.h>
  91 #include <machine/pcb.h>
  92 #ifdef SMP
  93 #include <machine/smp.h>
  94 #endif
  95 #include <machine/tss.h>
  96
  97 #ifdef KDTRACE_HOOKS
  98 #include <sys/dtrace_bsd.h>
  99
 100 /*
 101  * This is a hook which is initialised by the dtrace module
 102  * to handle traps which might occur during DTrace probe
 103  * execution.
 104  */
 105 dtrace_trap_func_t      dtrace_trap_func;
 106
 107 dtrace_doubletrap_func_t        dtrace_doubletrap_func;
 108
 109 /*
 110  * This is a hook which is initialised by the systrace module
 111  * when it is loaded. This keeps the DTrace syscall provider
 112  * implementation opaque.
 113  */
 114 systrace_probe_func_t   systrace_probe_func;
 115
 116 /*
 117  * These hooks are necessary for the pid, usdt and fasttrap providers.
 118  */
 119 dtrace_fasttrap_probe_ptr_t     dtrace_fasttrap_probe_ptr;
 120 dtrace_pid_probe_ptr_t          dtrace_pid_probe_ptr;
 121 dtrace_return_probe_ptr_t       dtrace_return_probe_ptr;
 122 #endif
 123
 124 extern void trap(struct trapframe *frame);
 125 extern void syscall(struct trapframe *frame);
 126 void dblfault_handler(struct trapframe *frame);
 127
 128 static int trap_pfault(struct trapframe *, int);
 129 static void trap_fatal(struct trapframe *, vm_offset_t);
 130
 131 #define MAX_TRAP_MSG            33
 132 static char *trap_msg[] = {
 133         "",                                     /*  0 unused */
 134         "privileged instruction fault",         /*  1 T_PRIVINFLT */
 135         "",                                     /*  2 unused */
 136         "breakpoint instruction fault",         /*  3 T_BPTFLT */
 137         "",                                     /*  4 unused */
 138         "",                                     /*  5 unused */
 139         "arithmetic trap",                      /*  6 T_ARITHTRAP */
 140         "",                                     /*  7 unused */
 141         "",                                     /*  8 unused */
 142         "general protection fault",             /*  9 T_PROTFLT */
 143         "trace trap",                           /* 10 T_TRCTRAP */
 144         "",                                     /* 11 unused */
 145         "page fault",                           /* 12 T_PAGEFLT */
 146         "",                                     /* 13 unused */
 147         "alignment fault",                      /* 14 T_ALIGNFLT */
 148         "",                                     /* 15 unused */
 149         "",                                     /* 16 unused */
 150         "",                                     /* 17 unused */
 151         "integer divide fault",                 /* 18 T_DIVIDE */
 152         "non-maskable interrupt trap",          /* 19 T_NMI */
 153         "overflow trap",                        /* 20 T_OFLOW */
 154         "FPU bounds check fault",               /* 21 T_BOUND */
 155         "FPU device not available",             /* 22 T_DNA */
 156         "double fault",                         /* 23 T_DOUBLEFLT */
 157         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
 158         "invalid TSS fault",                    /* 25 T_TSSFLT */
 159         "segment not present fault",            /* 26 T_SEGNPFLT */
 160         "stack fault",                          /* 27 T_STKFLT */
 161         "machine check trap",                   /* 28 T_MCHK */
 162         "SIMD floating-point exception",        /* 29 T_XMMFLT */
 163         "reserved (unknown) fault",             /* 30 T_RESERVED */
 164         "",                                     /* 31 unused (reserved) */
 165         "DTrace pid return trap",               /* 32 T_DTRACE_RET */
 166         "DTrace fasttrap probe trap",           /* 33 T_DTRACE_PROBE */
 167 };
 168
 169 #ifdef KDB
 170 static int kdb_on_nmi = 1;
 171 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
 172         &kdb_on_nmi, 0, "Go to KDB on NMI");
 173 TUNABLE_INT("machdep.kdb_on_nmi", &kdb_on_nmi);
 174 #endif
 175 static int panic_on_nmi = 1;
 176 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
 177         &panic_on_nmi, 0, "Panic on NMI");
 178 TUNABLE_INT("machdep.panic_on_nmi", &panic_on_nmi);
 179 static int prot_fault_translation = 0;
 180 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
 181         &prot_fault_translation, 0, "Select signal to deliver on protection fault");
 182
 183 /*
 184  * Exception, fault, and trap interface to the FreeBSD kernel.
 185  * This common code is called from assembly language IDT gate entry
 186  * routines that prepare a suitable stack frame, and restore this
 187  * frame after the exception has been processed.
 188  */
 189
 190 void
 191 trap(struct trapframe *frame)
 192 {
 193         struct thread *td = curthread;
 194         struct proc *p = td->td_proc;
 195         int i = 0, ucode = 0, code;
 196         u_int type;
 197         register_t addr = 0;
 198         ksiginfo_t ksi;
 199
 200         PCPU_INC(cnt.v_trap);
 201         type = frame->tf_trapno;
 202
 203 #ifdef SMP
 204         /* Handler for NMI IPIs used for stopping CPUs. */
 205         if (type == T_NMI) {
 206                  if (ipi_nmi_handler() == 0)
 207                            goto out;
 208         }
 209 #endif /* SMP */
 210
 211 #ifdef KDB
 212         if (kdb_active) {
 213                 kdb_reenter();
 214                 goto out;
 215         }
 216 #endif
 217
 218         if (type == T_RESERVED) {
 219                 trap_fatal(frame, 0);
 220                 goto out;
 221         }
 222
 223 #ifdef  HWPMC_HOOKS
 224         /*
 225          * CPU PMCs interrupt using an NMI.  If the PMC module is
 226          * active, pass the 'rip' value to the PMC module's interrupt
 227          * handler.  A return value of '1' from the handler means that
 228          * the NMI was handled by it and we can return immediately.
 229          */
 230         if (type == T_NMI && pmc_intr &&
 231             (*pmc_intr)(PCPU_GET(cpuid), frame))
 232                 goto out;
 233 #endif
 234
 235         if (type == T_MCHK) {
 236                 mca_intr();
 237                 goto out;
 238         }
 239
 240 #ifdef KDTRACE_HOOKS
 241         /*
 242          * A trap can occur while DTrace executes a probe. Before
 243          * executing the probe, DTrace blocks re-scheduling and sets
 244          * a flag in it's per-cpu flags to indicate that it doesn't
 245          * want to fault. On returning from the probe, the no-fault
 246          * flag is cleared and finally re-scheduling is enabled.
 247          *
 248          * If the DTrace kernel module has registered a trap handler,
 249          * call it and if it returns non-zero, assume that it has
 250          * handled the trap and modified the trap frame so that this
 251          * function can return normally.
 252          */
 253         if (type == T_DTRACE_PROBE || type == T_DTRACE_RET ||
 254             type == T_BPTFLT) {
 255                 struct reg regs;
 256
 257                 fill_frame_regs(frame, &regs);
 258                 if (type == T_DTRACE_PROBE &&
 259                     dtrace_fasttrap_probe_ptr != NULL &&
 260                     dtrace_fasttrap_probe_ptr(&regs) == 0)
 261                         goto out;
 262                 else if (type == T_BPTFLT &&
 263                     dtrace_pid_probe_ptr != NULL &&
 264                     dtrace_pid_probe_ptr(&regs) == 0)
 265                         goto out;
 266                 else if (type == T_DTRACE_RET &&
 267                     dtrace_return_probe_ptr != NULL &&
 268                     dtrace_return_probe_ptr(&regs) == 0)
 269                         goto out;
 270         }
 271         if (dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
 272                 goto out;
 273 #endif
 274
 275         if ((frame->tf_rflags & PSL_I) == 0) {
 276                 /*
 277                  * Buggy application or kernel code has disabled
 278                  * interrupts and then trapped.  Enabling interrupts
 279                  * now is wrong, but it is better than running with
 280                  * interrupts disabled until they are accidentally
 281                  * enabled later.
 282                  */
 283                 if (ISPL(frame->tf_cs) == SEL_UPL)
 284                         uprintf(
 285                             "pid %ld (%s): trap %d with interrupts disabled\n",
 286                             (long)curproc->p_pid, curthread->td_name, type);
 287                 else if (type != T_NMI && type != T_BPTFLT &&
 288                     type != T_TRCTRAP) {
 289                         /*
 290                          * XXX not quite right, since this may be for a
 291                          * multiple fault in user mode.
 292                          */
 293                         printf("kernel trap %d with interrupts disabled\n",
 294                             type);
 295
 296                         /*
 297                          * We shouldn't enable interrupts while holding a
 298                          * spin lock.
 299                          */
 300                         if (td->td_md.md_spinlock_count == 0)
 301                                 enable_intr();
 302                 }
 303         }
 304
 305         code = frame->tf_err;
 306
 307         if (ISPL(frame->tf_cs) == SEL_UPL) {
 308                 /* user trap */
 309
 310                 td->td_pticks = 0;
 311                 td->td_frame = frame;
 312                 addr = frame->tf_rip;
 313                 if (td->td_ucred != p->p_ucred)
 314                         cred_update_thread(td);
 315
 316                 switch (type) {
 317                 case T_PRIVINFLT:       /* privileged instruction fault */
 318                         i = SIGILL;
 319                         ucode = ILL_PRVOPC;
 320                         break;
 321
 322                 case T_BPTFLT:          /* bpt instruction fault */
 323                 case T_TRCTRAP:         /* trace trap */
 324                         enable_intr();
 325                         frame->tf_rflags &= ~PSL_T;
 326                         i = SIGTRAP;
 327                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 328                         break;
 329
 330                 case T_ARITHTRAP:       /* arithmetic trap */
 331                         ucode = fputrap();
 332                         if (ucode == -1)
 333                                 goto userout;
 334                         i = SIGFPE;
 335                         break;
 336
 337                 case T_PROTFLT:         /* general protection fault */
 338                         i = SIGBUS;
 339                         ucode = BUS_OBJERR;
 340                         break;
 341                 case T_STKFLT:          /* stack fault */
 342                 case T_SEGNPFLT:        /* segment not present fault */
 343                         i = SIGBUS;
 344                         ucode = BUS_ADRERR;
 345                         break;
 346                 case T_TSSFLT:          /* invalid TSS fault */
 347                         i = SIGBUS;
 348                         ucode = BUS_OBJERR;
 349                         break;
 350                 case T_DOUBLEFLT:       /* double fault */
 351                 default:
 352                         i = SIGBUS;
 353                         ucode = BUS_OBJERR;
 354                         break;
 355
 356                 case T_PAGEFLT:         /* page fault */
 357                         addr = frame->tf_addr;
 358                         i = trap_pfault(frame, TRUE);
 359                         if (i == -1)
 360                                 goto userout;
 361                         if (i == 0)
 362                                 goto user;
 363
 364                         if (i == SIGSEGV)
 365                                 ucode = SEGV_MAPERR;
 366                         else {
 367                                 if (prot_fault_translation == 0) {
 368                                         /*
 369                                          * Autodetect.
 370                                          * This check also covers the images
 371                                          * without the ABI-tag ELF note.
 372                                          */
 373                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 374                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 375                                                 i = SIGSEGV;
 376                                                 ucode = SEGV_ACCERR;
 377                                         } else {
 378                                                 i = SIGBUS;
 379                                                 ucode = BUS_PAGE_FAULT;
 380                                         }
 381                                 } else if (prot_fault_translation == 1) {
 382                                         /*
 383                                          * Always compat mode.
 384                                          */
 385                                         i = SIGBUS;
 386                                         ucode = BUS_PAGE_FAULT;
 387                                 } else {
 388                                         /*
 389                                          * Always SIGSEGV mode.
 390                                          */
 391                                         i = SIGSEGV;
 392                                         ucode = SEGV_ACCERR;
 393                                 }
 394                         }
 395                         break;
 396
 397                 case T_DIVIDE:          /* integer divide fault */
 398                         ucode = FPE_INTDIV;
 399                         i = SIGFPE;
 400                         break;
 401
 402 #ifdef DEV_ISA
 403                 case T_NMI:
 404                         /* machine/parity/power fail/"kitchen sink" faults */
 405                         if (isa_nmi(code) == 0) {
 406 #ifdef KDB
 407                                 /*
 408                                  * NMI can be hooked up to a pushbutton
 409                                  * for debugging.
 410                                  */
 411                                 if (kdb_on_nmi) {
 412                                         printf ("NMI ... going to debugger\n");
 413                                         kdb_trap(type, 0, frame);
 414                                 }
 415 #endif /* KDB */
 416                                 goto userout;
 417                         } else if (panic_on_nmi)
 418                                 panic("NMI indicates hardware failure");
 419                         break;
 420 #endif /* DEV_ISA */
 421
 422                 case T_OFLOW:           /* integer overflow fault */
 423                         ucode = FPE_INTOVF;
 424                         i = SIGFPE;
 425                         break;
 426
 427                 case T_BOUND:           /* bounds check fault */
 428                         ucode = FPE_FLTSUB;
 429                         i = SIGFPE;
 430                         break;
 431
 432                 case T_DNA:
 433                         /* transparent fault (due to context switch "late") */
 434                         KASSERT(PCB_USER_FPU(td->td_pcb),
 435                             ("kernel FPU ctx has leaked"));
 436                         fpudna();
 437                         goto userout;
 438
 439                 case T_FPOPFLT:         /* FPU operand fetch fault */
 440                         ucode = ILL_COPROC;
 441                         i = SIGILL;
 442                         break;
 443
 444                 case T_XMMFLT:          /* SIMD floating-point exception */
 445                         ucode = 0; /* XXX */
 446                         i = SIGFPE;
 447                         break;
 448                 }
 449         } else {
 450                 /* kernel trap */
 451
 452                 KASSERT(cold || td->td_ucred != NULL,
 453                     ("kernel trap doesn't have ucred"));
 454                 switch (type) {
 455                 case T_PAGEFLT:                 /* page fault */
 456                         (void) trap_pfault(frame, FALSE);
 457                         goto out;
 458
 459                 case T_DNA:
 460                         KASSERT(!PCB_USER_FPU(td->td_pcb),
 461                             ("Unregistered use of FPU in kernel"));
 462                         fpudna();
 463                         goto out;
 464
 465                 case T_ARITHTRAP:       /* arithmetic trap */
 466                 case T_XMMFLT:          /* SIMD floating-point exception */
 467                 case T_FPOPFLT:         /* FPU operand fetch fault */
 468                         /*
 469                          * XXXKIB for now disable any FPU traps in kernel
 470                          * handler registration seems to be overkill
 471                          */
 472                         trap_fatal(frame, 0);
 473                         goto out;
 474
 475                 case T_STKFLT:          /* stack fault */
 476                         break;
 477
 478                 case T_PROTFLT:         /* general protection fault */
 479                 case T_SEGNPFLT:        /* segment not present fault */
 480                         if (td->td_intr_nesting_level != 0)
 481                                 break;
 482
 483                         /*
 484                          * Invalid segment selectors and out of bounds
 485                          * %rip's and %rsp's can be set up in user mode.
 486                          * This causes a fault in kernel mode when the
 487                          * kernel tries to return to user mode.  We want
 488                          * to get this fault so that we can fix the
 489                          * problem here and not have to check all the
 490                          * selectors and pointers when the user changes
 491                          * them.
 492                          */
 493                         if (frame->tf_rip == (long)doreti_iret) {
 494                                 frame->tf_rip = (long)doreti_iret_fault;
 495                                 goto out;
 496                         }
 497                         if (frame->tf_rip == (long)ld_ds) {
 498                                 frame->tf_rip = (long)ds_load_fault;
 499                                 goto out;
 500                         }
 501                         if (frame->tf_rip == (long)ld_es) {
 502                                 frame->tf_rip = (long)es_load_fault;
 503                                 goto out;
 504                         }
 505                         if (frame->tf_rip == (long)ld_fs) {
 506                                 frame->tf_rip = (long)fs_load_fault;
 507                                 goto out;
 508                         }
 509                         if (frame->tf_rip == (long)ld_gs) {
 510                                 frame->tf_rip = (long)gs_load_fault;
 511                                 goto out;
 512                         }
 513                         if (frame->tf_rip == (long)ld_gsbase) {
 514                                 frame->tf_rip = (long)gsbase_load_fault;
 515                                 goto out;
 516                         }
 517                         if (frame->tf_rip == (long)ld_fsbase) {
 518                                 frame->tf_rip = (long)fsbase_load_fault;
 519                                 goto out;
 520                         }
 521                         if (PCPU_GET(curpcb)->pcb_onfault != NULL) {
 522                                 frame->tf_rip =
 523                                     (long)PCPU_GET(curpcb)->pcb_onfault;
 524                                 goto out;
 525                         }
 526                         break;
 527
 528                 case T_TSSFLT:
 529                         /*
 530                          * PSL_NT can be set in user mode and isn't cleared
 531                          * automatically when the kernel is entered.  This
 532                          * causes a TSS fault when the kernel attempts to
 533                          * `iret' because the TSS link is uninitialized.  We
 534                          * want to get this fault so that we can fix the
 535                          * problem here and not every time the kernel is
 536                          * entered.
 537                          */
 538                         if (frame->tf_rflags & PSL_NT) {
 539                                 frame->tf_rflags &= ~PSL_NT;
 540                                 goto out;
 541                         }
 542                         break;
 543
 544                 case T_TRCTRAP:  /* trace trap */
 545                         /*
 546                          * Ignore debug register trace traps due to
 547                          * accesses in the user's address space, which
 548                          * can happen under several conditions such as
 549                          * if a user sets a watchpoint on a buffer and
 550                          * then passes that buffer to a system call.
 551                          * We still want to get TRCTRAPS for addresses
 552                          * in kernel space because that is useful when
 553                          * debugging the kernel.
 554                          */
 555                         if (user_dbreg_trap()) {
 556                                 /*
 557                                  * Reset breakpoint bits because the
 558                                  * processor doesn't
 559                                  */
 560                                 /* XXX check upper bits here */
 561                                 load_dr6(rdr6() & 0xfffffff0);
 562                                 goto out;
 563                         }
 564                         /*
 565                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 566                          */
 567                 case T_BPTFLT:
 568                         /*
 569                          * If KDB is enabled, let it handle the debugger trap.
 570                          * Otherwise, debugger traps "can't happen".
 571                          */
 572 #ifdef KDB
 573                         if (kdb_trap(type, 0, frame))
 574                                 goto out;
 575 #endif
 576                         break;
 577
 578 #ifdef DEV_ISA
 579                 case T_NMI:
 580                         /* machine/parity/power fail/"kitchen sink" faults */
 581                         if (isa_nmi(code) == 0) {
 582 #ifdef KDB
 583                                 /*
 584                                  * NMI can be hooked up to a pushbutton
 585                                  * for debugging.
 586                                  */
 587                                 if (kdb_on_nmi) {
 588                                         printf ("NMI ... going to debugger\n");
 589                                         kdb_trap(type, 0, frame);
 590                                 }
 591 #endif /* KDB */
 592                                 goto out;
 593                         } else if (panic_on_nmi == 0)
 594                                 goto out;
 595                         /* FALLTHROUGH */
 596 #endif /* DEV_ISA */
 597                 }
 598
 599                 trap_fatal(frame, 0);
 600                 goto out;
 601         }
 602
 603         /* Translate fault for emulators (e.g. Linux) */
 604         if (*p->p_sysent->sv_transtrap)
 605                 i = (*p->p_sysent->sv_transtrap)(i, type);
 606
 607         ksiginfo_init_trap(&ksi);
 608         ksi.ksi_signo = i;
 609         ksi.ksi_code = ucode;
 610         ksi.ksi_trapno = type;
 611         ksi.ksi_addr = (void *)addr;
 612         trapsignal(td, &ksi);
 613
 614 user:
 615         userret(td, frame);
 616         mtx_assert(&Giant, MA_NOTOWNED);
 617         KASSERT(PCB_USER_FPU(td->td_pcb),
 618             ("Return from trap with kernel FPU ctx leaked"));
 619 userout:
 620 out:
 621         return;
 622 }
 623
 624 static int
 625 trap_pfault(frame, usermode)
 626         struct trapframe *frame;
 627         int usermode;
 628 {
 629         vm_offset_t va;
 630         struct vmspace *vm = NULL;
 631         vm_map_t map;
 632         int rv = 0;
 633         vm_prot_t ftype;
 634         struct thread *td = curthread;
 635         struct proc *p = td->td_proc;
 636         vm_offset_t eva = frame->tf_addr;
 637
 638         if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
 639                 /*
 640                  * Due to both processor errata and lazy TLB invalidation when
 641                  * access restrictions are removed from virtual pages, memory
 642                  * accesses that are allowed by the physical mapping layer may
 643                  * nonetheless cause one spurious page fault per virtual page.
 644                  * When the thread is executing a "no faulting" section that
 645                  * is bracketed by vm_fault_{disable,enable}_pagefaults(),
 646                  * every page fault is treated as a spurious page fault,
 647                  * unless it accesses the same virtual address as the most
 648                  * recent page fault within the same "no faulting" section.
 649                  */
 650                 if (td->td_md.md_spurflt_addr != eva ||
 651                     (td->td_pflags & TDP_RESETSPUR) != 0) {
 652                         /*
 653                          * Do nothing to the TLB.  A stale TLB entry is
 654                          * flushed automatically by a page fault.
 655                          */
 656                         td->td_md.md_spurflt_addr = eva;
 657                         td->td_pflags &= ~TDP_RESETSPUR;
 658                         return (0);
 659                 }
 660         } else {
 661                 /*
 662                  * If we get a page fault while in a critical section, then
 663                  * it is most likely a fatal kernel page fault.  The kernel
 664                  * is already going to panic trying to get a sleep lock to
 665                  * do the VM lookup, so just consider it a fatal trap so the
 666                  * kernel can print out a useful trap message and even get
 667                  * to the debugger.
 668                  *
 669                  * If we get a page fault while holding a non-sleepable
 670                  * lock, then it is most likely a fatal kernel page fault.
 671                  * If WITNESS is enabled, then it's going to whine about
 672                  * bogus LORs with various VM locks, so just skip to the
 673                  * fatal trap handling directly.
 674                  */
 675                 if (td->td_critnest != 0 ||
 676                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 677                     "Kernel page fault") != 0) {
 678                         trap_fatal(frame, eva);
 679                         return (-1);
 680                 }
 681         }
 682         va = trunc_page(eva);
 683         if (va >= VM_MIN_KERNEL_ADDRESS) {
 684                 /*
 685                  * Don't allow user-mode faults in kernel address space.
 686                  */
 687                 if (usermode)
 688                         goto nogo;
 689
 690                 map = kernel_map;
 691         } else {
 692                 /*
 693                  * This is a fault on non-kernel virtual memory.
 694                  * vm is initialized above to NULL. If curproc is NULL
 695                  * or curproc->p_vmspace is NULL the fault is fatal.
 696                  */
 697                 if (p != NULL)
 698                         vm = p->p_vmspace;
 699
 700                 if (vm == NULL)
 701                         goto nogo;
 702
 703                 map = &vm->vm_map;
 704
 705                 /*
 706                  * When accessing a usermode address, kernel must be
 707                  * ready to accept the page fault, and provide a
 708                  * handling routine.  Since accessing the address
 709                  * without the handler is a bug, do not try to handle
 710                  * it normally, and panic immediately.
 711                  */
 712                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 713                     PCPU_GET(curpcb)->pcb_onfault == NULL)) {
 714                         trap_fatal(frame, eva);
 715                         return (-1);
 716                 }
 717         }
 718
 719         /*
 720          * PGEX_I is defined only if the execute disable bit capability is
 721          * supported and enabled.
 722          */
 723         if (frame->tf_err & PGEX_W)
 724                 ftype = VM_PROT_WRITE;
 725         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 726                 ftype = VM_PROT_EXECUTE;
 727         else
 728                 ftype = VM_PROT_READ;
 729
 730         if (map != kernel_map) {
 731                 /*
 732                  * Keep swapout from messing with us during this
 733                  *      critical time.
 734                  */
 735                 PROC_LOCK(p);
 736                 ++p->p_lock;
 737                 PROC_UNLOCK(p);
 738
 739                 /* Fault in the user page: */
 740                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 741
 742                 PROC_LOCK(p);
 743                 --p->p_lock;
 744                 PROC_UNLOCK(p);
 745         } else {
 746                 /*
 747                  * Don't have to worry about process locking or stacks in the
 748                  * kernel.
 749                  */
 750                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 751         }
 752         if (rv == KERN_SUCCESS) {
 753 #ifdef HWPMC_HOOKS
 754                 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
 755                         PMC_SOFT_CALL_TF( , , page_fault, all, frame);
 756                         if (ftype == VM_PROT_READ)
 757                                 PMC_SOFT_CALL_TF( , , page_fault, read,
 758                                     frame);
 759                         else
 760                                 PMC_SOFT_CALL_TF( , , page_fault, write,
 761                                     frame);
 762                 }
 763 #endif
 764                 return (0);
 765         }
 766 nogo:
 767         if (!usermode) {
 768                 if (td->td_intr_nesting_level == 0 &&
 769                     PCPU_GET(curpcb)->pcb_onfault != NULL) {
 770                         frame->tf_rip = (long)PCPU_GET(curpcb)->pcb_onfault;
 771                         return (0);
 772                 }
 773                 trap_fatal(frame, eva);
 774                 return (-1);
 775         }
 776
 777         return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 778 }
 779
 780 static void
 781 trap_fatal(frame, eva)
 782         struct trapframe *frame;
 783         vm_offset_t eva;
 784 {
 785         int code, ss;
 786         u_int type;
 787         long esp;
 788         struct soft_segment_descriptor softseg;
 789         char *msg;
 790
 791         code = frame->tf_err;
 792         type = frame->tf_trapno;
 793         sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)],
 794             &softseg);
 795
 796         if (type <= MAX_TRAP_MSG)
 797                 msg = trap_msg[type];
 798         else
 799                 msg = "UNKNOWN";
 800         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 801             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
 802 #ifdef SMP
 803         /* two separate prints in case of a trap on an unmapped page */
 804         printf("cpuid = %d; ", PCPU_GET(cpuid));
 805         printf("apic id = %02x\n", PCPU_GET(apic_id));
 806 #endif
 807         if (type == T_PAGEFLT) {
 808                 printf("fault virtual address   = 0x%lx\n", eva);
 809                 printf("fault code              = %s %s %s, %s\n",
 810                         code & PGEX_U ? "user" : "supervisor",
 811                         code & PGEX_W ? "write" : "read",
 812                         code & PGEX_I ? "instruction" : "data",
 813                         code & PGEX_P ? "protection violation" : "page not present");
 814         }
 815         printf("instruction pointer     = 0x%lx:0x%lx\n",
 816                frame->tf_cs & 0xffff, frame->tf_rip);
 817         if (ISPL(frame->tf_cs) == SEL_UPL) {
 818                 ss = frame->tf_ss & 0xffff;
 819                 esp = frame->tf_rsp;
 820         } else {
 821                 ss = GSEL(GDATA_SEL, SEL_KPL);
 822                 esp = (long)&frame->tf_rsp;
 823         }
 824         printf("stack pointer           = 0x%x:0x%lx\n", ss, esp);
 825         printf("frame pointer           = 0x%x:0x%lx\n", ss, frame->tf_rbp);
 826         printf("code segment            = base 0x%lx, limit 0x%lx, type 0x%x\n",
 827                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 828         printf("                        = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
 829                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
 830                softseg.ssd_gran);
 831         printf("processor eflags        = ");
 832         if (frame->tf_rflags & PSL_T)
 833                 printf("trace trap, ");
 834         if (frame->tf_rflags & PSL_I)
 835                 printf("interrupt enabled, ");
 836         if (frame->tf_rflags & PSL_NT)
 837                 printf("nested task, ");
 838         if (frame->tf_rflags & PSL_RF)
 839                 printf("resume, ");
 840         printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
 841         printf("current process         = ");
 842         if (curproc) {
 843                 printf("%lu (%s)\n",
 844                     (u_long)curproc->p_pid, curthread->td_name ?
 845                     curthread->td_name : "");
 846         } else {
 847                 printf("Idle\n");
 848         }
 849
 850 #ifdef KDB
 851         if (debugger_on_panic || kdb_active)
 852                 if (kdb_trap(type, 0, frame))
 853                         return;
 854 #endif
 855         printf("trap number             = %d\n", type);
 856         if (type <= MAX_TRAP_MSG)
 857                 panic("%s", trap_msg[type]);
 858         else
 859                 panic("unknown/reserved trap");
 860 }
 861
 862 /*
 863  * Double fault handler. Called when a fault occurs while writing
 864  * a frame for a trap/exception onto the stack. This usually occurs
 865  * when the stack overflows (such is the case with infinite recursion,
 866  * for example).
 867  */
 868 void
 869 dblfault_handler(struct trapframe *frame)
 870 {
 871 #ifdef KDTRACE_HOOKS
 872         if (dtrace_doubletrap_func != NULL)
 873                 (*dtrace_doubletrap_func)();
 874 #endif
 875         printf("\nFatal double fault\n");
 876         printf("rip = 0x%lx\n", frame->tf_rip);
 877         printf("rsp = 0x%lx\n", frame->tf_rsp);
 878         printf("rbp = 0x%lx\n", frame->tf_rbp);
 879 #ifdef SMP
 880         /* two separate prints in case of a trap on an unmapped page */
 881         printf("cpuid = %d; ", PCPU_GET(cpuid));
 882         printf("apic id = %02x\n", PCPU_GET(apic_id));
 883 #endif
 884         panic("double fault");
 885 }
 886
 887 int
 888 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
 889 {
 890         struct proc *p;
 891         struct trapframe *frame;
 892         register_t *argp;
 893         caddr_t params;
 894         int reg, regcnt, error;
 895
 896         p = td->td_proc;
 897         frame = td->td_frame;
 898         reg = 0;
 899         regcnt = 6;
 900
 901         params = (caddr_t)frame->tf_rsp + sizeof(register_t);
 902         sa->code = frame->tf_rax;
 903
 904         if (sa->code == SYS_syscall || sa->code == SYS___syscall) {
 905                 sa->code = frame->tf_rdi;
 906                 reg++;
 907                 regcnt--;
 908         }
 909         if (p->p_sysent->sv_mask)
 910                 sa->code &= p->p_sysent->sv_mask;
 911
 912         if (sa->code >= p->p_sysent->sv_size)
 913                 sa->callp = &p->p_sysent->sv_table[0];
 914         else
 915                 sa->callp = &p->p_sysent->sv_table[sa->code];
 916
 917         sa->narg = sa->callp->sy_narg;
 918         KASSERT(sa->narg <= sizeof(sa->args) / sizeof(sa->args[0]),
 919             ("Too many syscall arguments!"));
 920         error = 0;
 921         argp = &frame->tf_rdi;
 922         argp += reg;
 923         bcopy(argp, sa->args, sizeof(sa->args[0]) * regcnt);
 924         if (sa->narg > regcnt) {
 925                 KASSERT(params != NULL, ("copyin args with no params!"));
 926                 error = copyin(params, &sa->args[regcnt],
 927                     (sa->narg - regcnt) * sizeof(sa->args[0]));
 928         }
 929
 930         if (error == 0) {
 931                 td->td_retval[0] = 0;
 932                 td->td_retval[1] = frame->tf_rdx;
 933         }
 934
 935         return (error);
 936 }
 937
 938 #include "../../kern/subr_syscall.c"
 939
 940 /*
 941  * System call handler for native binaries.  The trap frame is already
 942  * set up by the assembler trampoline and a pointer to it is saved in
 943  * td_frame.
 944  */
 945 void
 946 amd64_syscall(struct thread *td, int traced)
 947 {
 948         struct syscall_args sa;
 949         int error;
 950         ksiginfo_t ksi;
 951
 952 #ifdef DIAGNOSTIC
 953         if (ISPL(td->td_frame->tf_cs) != SEL_UPL) {
 954                 panic("syscall");
 955                 /* NOT REACHED */
 956         }
 957 #endif
 958         error = syscallenter(td, &sa);
 959
 960         /*
 961          * Traced syscall.
 962          */
 963         if (__predict_false(traced)) {
 964                 td->td_frame->tf_rflags &= ~PSL_T;
 965                 ksiginfo_init_trap(&ksi);
 966                 ksi.ksi_signo = SIGTRAP;
 967                 ksi.ksi_code = TRAP_TRACE;
 968                 ksi.ksi_addr = (void *)td->td_frame->tf_rip;
 969                 trapsignal(td, &ksi);
 970         }
 971
 972         KASSERT(PCB_USER_FPU(td->td_pcb),
 973             ("System call %s returing with kernel FPU ctx leaked",
 974              syscallname(td->td_proc, sa.code)));
 975         KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td),
 976             ("System call %s returning with mangled pcb_save",
 977              syscallname(td->td_proc, sa.code)));
 978
 979         syscallret(td, error, &sa);
 980 }