sys/amd64/amd64/trap.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-4-Clause
   3  *
   4  * Copyright (C) 1994, David Greenman
   5  * Copyright (c) 1990, 1993
   6  *      The Regents of the University of California.  All rights reserved.
   7  *
   8  * This code is derived from software contributed to Berkeley by
   9  * the University of Utah, and William Jolitz.
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  * 3. All advertising materials mentioning features or use of this software
  20  *    must display the following acknowledgement:
  21  *      This product includes software developed by the University of
  22  *      California, Berkeley and its contributors.
  23  * 4. Neither the name of the University nor the names of its contributors
  24  *    may be used to endorse or promote products derived from this software
  25  *    without specific prior written permission.
  26  *
  27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  37  * SUCH DAMAGE.
  38  *
  39  *      from: @(#)trap.c        7.4 (Berkeley) 5/13/91
  40  */
  41
  42 #include <sys/cdefs.h>
  43 __FBSDID("$FreeBSD$");
  44
  45 /*
  46  * AMD64 Trap and System call handling
  47  */
  48
  49 #include "opt_clock.h"
  50 #include "opt_cpu.h"
  51 #include "opt_hwpmc_hooks.h"
  52 #include "opt_isa.h"
  53 #include "opt_kdb.h"
  54 #include "opt_stack.h"
  55
  56 #include <sys/param.h>
  57 #include <sys/bus.h>
  58 #include <sys/systm.h>
  59 #include <sys/proc.h>
  60 #include <sys/pioctl.h>
  61 #include <sys/ptrace.h>
  62 #include <sys/kdb.h>
  63 #include <sys/kernel.h>
  64 #include <sys/ktr.h>
  65 #include <sys/lock.h>
  66 #include <sys/mutex.h>
  67 #include <sys/resourcevar.h>
  68 #include <sys/signalvar.h>
  69 #include <sys/syscall.h>
  70 #include <sys/sysctl.h>
  71 #include <sys/sysent.h>
  72 #include <sys/uio.h>
  73 #include <sys/vmmeter.h>
  74 #ifdef HWPMC_HOOKS
  75 #include <sys/pmckern.h>
  76 PMC_SOFT_DEFINE( , , page_fault, all);
  77 PMC_SOFT_DEFINE( , , page_fault, read);
  78 PMC_SOFT_DEFINE( , , page_fault, write);
  79 #endif
  80
  81 #include <vm/vm.h>
  82 #include <vm/vm_param.h>
  83 #include <vm/pmap.h>
  84 #include <vm/vm_kern.h>
  85 #include <vm/vm_map.h>
  86 #include <vm/vm_page.h>
  87 #include <vm/vm_extern.h>
  88
  89 #include <machine/cpu.h>
  90 #include <machine/intr_machdep.h>
  91 #include <x86/mca.h>
  92 #include <machine/md_var.h>
  93 #include <machine/pcb.h>
  94 #ifdef SMP
  95 #include <machine/smp.h>
  96 #endif
  97 #include <machine/stack.h>
  98 #include <machine/tss.h>
  99
 100 #ifdef KDTRACE_HOOKS
 101 #include <sys/dtrace_bsd.h>
 102 #endif
 103
 104 void __noinline trap(struct trapframe *frame);
 105 void trap_check(struct trapframe *frame);
 106 void dblfault_handler(struct trapframe *frame);
 107
 108 static int trap_pfault(struct trapframe *, int);
 109 static void trap_fatal(struct trapframe *, vm_offset_t);
 110
 111 #define MAX_TRAP_MSG            32
 112 static char *trap_msg[] = {
 113         "",                                     /*  0 unused */
 114         "privileged instruction fault",         /*  1 T_PRIVINFLT */
 115         "",                                     /*  2 unused */
 116         "breakpoint instruction fault",         /*  3 T_BPTFLT */
 117         "",                                     /*  4 unused */
 118         "",                                     /*  5 unused */
 119         "arithmetic trap",                      /*  6 T_ARITHTRAP */
 120         "",                                     /*  7 unused */
 121         "",                                     /*  8 unused */
 122         "general protection fault",             /*  9 T_PROTFLT */
 123         "trace trap",                           /* 10 T_TRCTRAP */
 124         "",                                     /* 11 unused */
 125         "page fault",                           /* 12 T_PAGEFLT */
 126         "",                                     /* 13 unused */
 127         "alignment fault",                      /* 14 T_ALIGNFLT */
 128         "",                                     /* 15 unused */
 129         "",                                     /* 16 unused */
 130         "",                                     /* 17 unused */
 131         "integer divide fault",                 /* 18 T_DIVIDE */
 132         "non-maskable interrupt trap",          /* 19 T_NMI */
 133         "overflow trap",                        /* 20 T_OFLOW */
 134         "FPU bounds check fault",               /* 21 T_BOUND */
 135         "FPU device not available",             /* 22 T_DNA */
 136         "double fault",                         /* 23 T_DOUBLEFLT */
 137         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
 138         "invalid TSS fault",                    /* 25 T_TSSFLT */
 139         "segment not present fault",            /* 26 T_SEGNPFLT */
 140         "stack fault",                          /* 27 T_STKFLT */
 141         "machine check trap",                   /* 28 T_MCHK */
 142         "SIMD floating-point exception",        /* 29 T_XMMFLT */
 143         "reserved (unknown) fault",             /* 30 T_RESERVED */
 144         "",                                     /* 31 unused (reserved) */
 145         "DTrace pid return trap",               /* 32 T_DTRACE_RET */
 146 };
 147
 148 static int prot_fault_translation;
 149 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RWTUN,
 150     &prot_fault_translation, 0,
 151     "Select signal to deliver on protection fault");
 152 static int uprintf_signal;
 153 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RWTUN,
 154     &uprintf_signal, 0,
 155     "Print debugging information on trap signal to ctty");
 156
 157 /*
 158  * Exception, fault, and trap interface to the FreeBSD kernel.
 159  * This common code is called from assembly language IDT gate entry
 160  * routines that prepare a suitable stack frame, and restore this
 161  * frame after the exception has been processed.
 162  */
 163
 164 void
 165 trap(struct trapframe *frame)
 166 {
 167         ksiginfo_t ksi;
 168         struct thread *td;
 169         struct proc *p;
 170         register_t addr;
 171 #ifdef KDB
 172         register_t dr6;
 173 #endif
 174         int signo, ucode;
 175         u_int type;
 176
 177         td = curthread;
 178         p = td->td_proc;
 179         signo = 0;
 180         ucode = 0;
 181         addr = 0;
 182
 183         VM_CNT_INC(v_trap);
 184         type = frame->tf_trapno;
 185
 186 #ifdef SMP
 187         /* Handler for NMI IPIs used for stopping CPUs. */
 188         if (type == T_NMI && ipi_nmi_handler() == 0)
 189                 return;
 190 #endif
 191
 192 #ifdef KDB
 193         if (kdb_active) {
 194                 kdb_reenter();
 195                 return;
 196         }
 197 #endif
 198
 199         if (type == T_RESERVED) {
 200                 trap_fatal(frame, 0);
 201                 return;
 202         }
 203
 204         if (type == T_NMI) {
 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 non-zero return value from the handler means that
 210                  * the NMI was consumed by it and we can return immediately.
 211                  */
 212                 if (pmc_intr != NULL &&
 213                     (*pmc_intr)(PCPU_GET(cpuid), frame) != 0)
 214                         return;
 215 #endif
 216
 217 #ifdef STACK
 218                 if (stack_nmi_handler(frame) != 0)
 219                         return;
 220 #endif
 221         }
 222
 223         if (type == T_MCHK) {
 224                 mca_intr();
 225                 return;
 226         }
 227
 228         if ((frame->tf_rflags & PSL_I) == 0) {
 229                 /*
 230                  * Buggy application or kernel code has disabled
 231                  * interrupts and then trapped.  Enabling interrupts
 232                  * now is wrong, but it is better than running with
 233                  * interrupts disabled until they are accidentally
 234                  * enabled later.
 235                  */
 236                 if (TRAPF_USERMODE(frame))
 237                         uprintf(
 238                             "pid %ld (%s): trap %d with interrupts disabled\n",
 239                             (long)curproc->p_pid, curthread->td_name, type);
 240                 else if (type != T_NMI && type != T_BPTFLT &&
 241                     type != T_TRCTRAP) {
 242                         /*
 243                          * XXX not quite right, since this may be for a
 244                          * multiple fault in user mode.
 245                          */
 246                         printf("kernel trap %d with interrupts disabled\n",
 247                             type);
 248
 249                         /*
 250                          * We shouldn't enable interrupts while holding a
 251                          * spin lock.
 252                          */
 253                         if (td->td_md.md_spinlock_count == 0)
 254                                 enable_intr();
 255                 }
 256         }
 257
 258         if (TRAPF_USERMODE(frame)) {
 259                 /* user trap */
 260
 261                 td->td_pticks = 0;
 262                 td->td_frame = frame;
 263                 addr = frame->tf_rip;
 264                 if (td->td_cowgen != p->p_cowgen)
 265                         thread_cow_update(td);
 266
 267                 switch (type) {
 268                 case T_PRIVINFLT:       /* privileged instruction fault */
 269                         signo = SIGILL;
 270                         ucode = ILL_PRVOPC;
 271                         break;
 272
 273                 case T_BPTFLT:          /* bpt instruction fault */
 274                 case T_TRCTRAP:         /* trace trap */
 275                         enable_intr();
 276 #ifdef KDTRACE_HOOKS
 277                         if (type == T_BPTFLT) {
 278                                 if (dtrace_pid_probe_ptr != NULL &&
 279                                     dtrace_pid_probe_ptr(frame) == 0)
 280                                         return;
 281                         }
 282 #endif
 283                         frame->tf_rflags &= ~PSL_T;
 284                         signo = SIGTRAP;
 285                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 286                         break;
 287
 288                 case T_ARITHTRAP:       /* arithmetic trap */
 289                         ucode = fputrap_x87();
 290                         if (ucode == -1)
 291                                 return;
 292                         signo = SIGFPE;
 293                         break;
 294
 295                 case T_PROTFLT:         /* general protection fault */
 296                         signo = SIGBUS;
 297                         ucode = BUS_OBJERR;
 298                         break;
 299                 case T_STKFLT:          /* stack fault */
 300                 case T_SEGNPFLT:        /* segment not present fault */
 301                         signo = SIGBUS;
 302                         ucode = BUS_ADRERR;
 303                         break;
 304                 case T_TSSFLT:          /* invalid TSS fault */
 305                         signo = SIGBUS;
 306                         ucode = BUS_OBJERR;
 307                         break;
 308                 case T_ALIGNFLT:
 309                         signo = SIGBUS;
 310                         ucode = BUS_ADRALN;
 311                         break;
 312                 case T_DOUBLEFLT:       /* double fault */
 313                 default:
 314                         signo = SIGBUS;
 315                         ucode = BUS_OBJERR;
 316                         break;
 317
 318                 case T_PAGEFLT:         /* page fault */
 319                         /*
 320                          * Emulator can take care about this trap?
 321                          */
 322                         if (*p->p_sysent->sv_trap != NULL &&
 323                             (*p->p_sysent->sv_trap)(td) == 0)
 324                                 return;
 325
 326                         addr = frame->tf_addr;
 327                         signo = trap_pfault(frame, TRUE);
 328                         if (signo == -1)
 329                                 return;
 330                         if (signo == 0)
 331                                 goto userret;
 332                         if (signo == SIGSEGV) {
 333                                 ucode = SEGV_MAPERR;
 334                         } else if (prot_fault_translation == 0) {
 335                                 /*
 336                                  * Autodetect.  This check also covers
 337                                  * the images without the ABI-tag ELF
 338                                  * note.
 339                                  */
 340                                 if (SV_CURPROC_ABI() == SV_ABI_FREEBSD &&
 341                                     p->p_osrel >= P_OSREL_SIGSEGV) {
 342                                         signo = SIGSEGV;
 343                                         ucode = SEGV_ACCERR;
 344                                 } else {
 345                                         signo = SIGBUS;
 346                                         ucode = BUS_PAGE_FAULT;
 347                                 }
 348                         } else if (prot_fault_translation == 1) {
 349                                 /*
 350                                  * Always compat mode.
 351                                  */
 352                                 signo = SIGBUS;
 353                                 ucode = BUS_PAGE_FAULT;
 354                         } else {
 355                                 /*
 356                                  * Always SIGSEGV mode.
 357                                  */
 358                                 signo = SIGSEGV;
 359                                 ucode = SEGV_ACCERR;
 360                         }
 361                         break;
 362
 363                 case T_DIVIDE:          /* integer divide fault */
 364                         ucode = FPE_INTDIV;
 365                         signo = SIGFPE;
 366                         break;
 367
 368 #ifdef DEV_ISA
 369                 case T_NMI:
 370                         nmi_handle_intr(type, frame);
 371                         return;
 372 #endif
 373
 374                 case T_OFLOW:           /* integer overflow fault */
 375                         ucode = FPE_INTOVF;
 376                         signo = SIGFPE;
 377                         break;
 378
 379                 case T_BOUND:           /* bounds check fault */
 380                         ucode = FPE_FLTSUB;
 381                         signo = SIGFPE;
 382                         break;
 383
 384                 case T_DNA:
 385                         /* transparent fault (due to context switch "late") */
 386                         KASSERT(PCB_USER_FPU(td->td_pcb),
 387                             ("kernel FPU ctx has leaked"));
 388                         fpudna();
 389                         return;
 390
 391                 case T_FPOPFLT:         /* FPU operand fetch fault */
 392                         ucode = ILL_COPROC;
 393                         signo = SIGILL;
 394                         break;
 395
 396                 case T_XMMFLT:          /* SIMD floating-point exception */
 397                         ucode = fputrap_sse();
 398                         if (ucode == -1)
 399                                 return;
 400                         signo = SIGFPE;
 401                         break;
 402 #ifdef KDTRACE_HOOKS
 403                 case T_DTRACE_RET:
 404                         enable_intr();
 405                         if (dtrace_return_probe_ptr != NULL)
 406                                 dtrace_return_probe_ptr(frame);
 407                         return;
 408 #endif
 409                 }
 410         } else {
 411                 /* kernel trap */
 412
 413                 KASSERT(cold || td->td_ucred != NULL,
 414                     ("kernel trap doesn't have ucred"));
 415                 switch (type) {
 416                 case T_PAGEFLT:                 /* page fault */
 417                         (void) trap_pfault(frame, FALSE);
 418                         return;
 419
 420                 case T_DNA:
 421                         if (PCB_USER_FPU(td->td_pcb))
 422                                 panic("Unregistered use of FPU in kernel");
 423                         fpudna();
 424                         return;
 425
 426                 case T_ARITHTRAP:       /* arithmetic trap */
 427                 case T_XMMFLT:          /* SIMD floating-point exception */
 428                 case T_FPOPFLT:         /* FPU operand fetch fault */
 429                         /*
 430                          * For now, supporting kernel handler
 431                          * registration for FPU traps is overkill.
 432                          */
 433                         trap_fatal(frame, 0);
 434                         return;
 435
 436                 case T_STKFLT:          /* stack fault */
 437                 case T_PROTFLT:         /* general protection fault */
 438                 case T_SEGNPFLT:        /* segment not present fault */
 439                         if (td->td_intr_nesting_level != 0)
 440                                 break;
 441
 442                         /*
 443                          * Invalid segment selectors and out of bounds
 444                          * %rip's and %rsp's can be set up in user mode.
 445                          * This causes a fault in kernel mode when the
 446                          * kernel tries to return to user mode.  We want
 447                          * to get this fault so that we can fix the
 448                          * problem here and not have to check all the
 449                          * selectors and pointers when the user changes
 450                          * them.
 451                          */
 452                         if (frame->tf_rip == (long)doreti_iret) {
 453                                 frame->tf_rip = (long)doreti_iret_fault;
 454                                 return;
 455                         }
 456                         if (frame->tf_rip == (long)ld_ds) {
 457                                 frame->tf_rip = (long)ds_load_fault;
 458                                 return;
 459                         }
 460                         if (frame->tf_rip == (long)ld_es) {
 461                                 frame->tf_rip = (long)es_load_fault;
 462                                 return;
 463                         }
 464                         if (frame->tf_rip == (long)ld_fs) {
 465                                 frame->tf_rip = (long)fs_load_fault;
 466                                 return;
 467                         }
 468                         if (frame->tf_rip == (long)ld_gs) {
 469                                 frame->tf_rip = (long)gs_load_fault;
 470                                 return;
 471                         }
 472                         if (frame->tf_rip == (long)ld_gsbase) {
 473                                 frame->tf_rip = (long)gsbase_load_fault;
 474                                 return;
 475                         }
 476                         if (frame->tf_rip == (long)ld_fsbase) {
 477                                 frame->tf_rip = (long)fsbase_load_fault;
 478                                 return;
 479                         }
 480                         if (curpcb->pcb_onfault != NULL) {
 481                                 frame->tf_rip = (long)curpcb->pcb_onfault;
 482                                 return;
 483                         }
 484                         break;
 485
 486                 case T_TSSFLT:
 487                         /*
 488                          * PSL_NT can be set in user mode and isn't cleared
 489                          * automatically when the kernel is entered.  This
 490                          * causes a TSS fault when the kernel attempts to
 491                          * `iret' because the TSS link is uninitialized.  We
 492                          * want to get this fault so that we can fix the
 493                          * problem here and not every time the kernel is
 494                          * entered.
 495                          */
 496                         if (frame->tf_rflags & PSL_NT) {
 497                                 frame->tf_rflags &= ~PSL_NT;
 498                                 return;
 499                         }
 500                         break;
 501
 502                 case T_TRCTRAP:  /* trace trap */
 503                         /*
 504                          * Ignore debug register trace traps due to
 505                          * accesses in the user's address space, which
 506                          * can happen under several conditions such as
 507                          * if a user sets a watchpoint on a buffer and
 508                          * then passes that buffer to a system call.
 509                          * We still want to get TRCTRAPS for addresses
 510                          * in kernel space because that is useful when
 511                          * debugging the kernel.
 512                          */
 513                         if (user_dbreg_trap()) {
 514                                 /*
 515                                  * Reset breakpoint bits because the
 516                                  * processor doesn't
 517                                  */
 518                                 load_dr6(rdr6() & ~0xf);
 519                                 return;
 520                         }
 521                         /*
 522                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 523                          */
 524                 case T_BPTFLT:
 525                         /*
 526                          * If KDB is enabled, let it handle the debugger trap.
 527                          * Otherwise, debugger traps "can't happen".
 528                          */
 529 #ifdef KDB
 530                         /* XXX %dr6 is not quite reentrant. */
 531                         dr6 = rdr6();
 532                         load_dr6(dr6 & ~0x4000);
 533                         if (kdb_trap(type, dr6, frame))
 534                                 return;
 535 #endif
 536                         break;
 537
 538 #ifdef DEV_ISA
 539                 case T_NMI:
 540                         nmi_handle_intr(type, frame);
 541                         return;
 542 #endif
 543                 }
 544
 545                 trap_fatal(frame, 0);
 546                 return;
 547         }
 548
 549         /* Translate fault for emulators (e.g. Linux) */
 550         if (*p->p_sysent->sv_transtrap != NULL)
 551                 signo = (*p->p_sysent->sv_transtrap)(signo, type);
 552
 553         ksiginfo_init_trap(&ksi);
 554         ksi.ksi_signo = signo;
 555         ksi.ksi_code = ucode;
 556         ksi.ksi_trapno = type;
 557         ksi.ksi_addr = (void *)addr;
 558         if (uprintf_signal) {
 559                 uprintf("pid %d comm %s: signal %d err %lx code %d type %d "
 560                     "addr 0x%lx rsp 0x%lx rip 0x%lx "
 561                     "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
 562                     p->p_pid, p->p_comm, signo, frame->tf_err, ucode, type,
 563                     addr, frame->tf_rsp, frame->tf_rip,
 564                     fubyte((void *)(frame->tf_rip + 0)),
 565                     fubyte((void *)(frame->tf_rip + 1)),
 566                     fubyte((void *)(frame->tf_rip + 2)),
 567                     fubyte((void *)(frame->tf_rip + 3)),
 568                     fubyte((void *)(frame->tf_rip + 4)),
 569                     fubyte((void *)(frame->tf_rip + 5)),
 570                     fubyte((void *)(frame->tf_rip + 6)),
 571                     fubyte((void *)(frame->tf_rip + 7)));
 572         }
 573         KASSERT((read_rflags() & PSL_I) != 0, ("interrupts disabled"));
 574         trapsignal(td, &ksi);
 575 userret:
 576         userret(td, frame);
 577         KASSERT(PCB_USER_FPU(td->td_pcb),
 578             ("Return from trap with kernel FPU ctx leaked"));
 579 }
 580
 581 /*
 582  * Ensure that we ignore any DTrace-induced faults. This function cannot
 583  * be instrumented, so it cannot generate such faults itself.
 584  */
 585 void
 586 trap_check(struct trapframe *frame)
 587 {
 588
 589 #ifdef KDTRACE_HOOKS
 590         if (dtrace_trap_func != NULL &&
 591             (*dtrace_trap_func)(frame, frame->tf_trapno) != 0)
 592                 return;
 593 #endif
 594         trap(frame);
 595 }
 596
 597 static int
 598 trap_pfault(struct trapframe *frame, int usermode)
 599 {
 600         struct thread *td;
 601         struct proc *p;
 602         vm_map_t map;
 603         vm_offset_t va;
 604         int rv;
 605         vm_prot_t ftype;
 606         vm_offset_t eva;
 607
 608         td = curthread;
 609         p = td->td_proc;
 610         eva = frame->tf_addr;
 611
 612         if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
 613                 /*
 614                  * Due to both processor errata and lazy TLB invalidation when
 615                  * access restrictions are removed from virtual pages, memory
 616                  * accesses that are allowed by the physical mapping layer may
 617                  * nonetheless cause one spurious page fault per virtual page.
 618                  * When the thread is executing a "no faulting" section that
 619                  * is bracketed by vm_fault_{disable,enable}_pagefaults(),
 620                  * every page fault is treated as a spurious page fault,
 621                  * unless it accesses the same virtual address as the most
 622                  * recent page fault within the same "no faulting" section.
 623                  */
 624                 if (td->td_md.md_spurflt_addr != eva ||
 625                     (td->td_pflags & TDP_RESETSPUR) != 0) {
 626                         /*
 627                          * Do nothing to the TLB.  A stale TLB entry is
 628                          * flushed automatically by a page fault.
 629                          */
 630                         td->td_md.md_spurflt_addr = eva;
 631                         td->td_pflags &= ~TDP_RESETSPUR;
 632                         return (0);
 633                 }
 634         } else {
 635                 /*
 636                  * If we get a page fault while in a critical section, then
 637                  * it is most likely a fatal kernel page fault.  The kernel
 638                  * is already going to panic trying to get a sleep lock to
 639                  * do the VM lookup, so just consider it a fatal trap so the
 640                  * kernel can print out a useful trap message and even get
 641                  * to the debugger.
 642                  *
 643                  * If we get a page fault while holding a non-sleepable
 644                  * lock, then it is most likely a fatal kernel page fault.
 645                  * If WITNESS is enabled, then it's going to whine about
 646                  * bogus LORs with various VM locks, so just skip to the
 647                  * fatal trap handling directly.
 648                  */
 649                 if (td->td_critnest != 0 ||
 650                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 651                     "Kernel page fault") != 0) {
 652                         trap_fatal(frame, eva);
 653                         return (-1);
 654                 }
 655         }
 656         va = trunc_page(eva);
 657         if (va >= VM_MIN_KERNEL_ADDRESS) {
 658                 /*
 659                  * Don't allow user-mode faults in kernel address space.
 660                  */
 661                 if (usermode)
 662                         return (SIGSEGV);
 663
 664                 map = kernel_map;
 665         } else {
 666                 map = &p->p_vmspace->vm_map;
 667
 668                 /*
 669                  * When accessing a usermode address, kernel must be
 670                  * ready to accept the page fault, and provide a
 671                  * handling routine.  Since accessing the address
 672                  * without the handler is a bug, do not try to handle
 673                  * it normally, and panic immediately.
 674                  */
 675                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 676                     curpcb->pcb_onfault == NULL)) {
 677                         trap_fatal(frame, eva);
 678                         return (-1);
 679                 }
 680         }
 681
 682         /*
 683          * If the trap was caused by errant bits in the PTE then panic.
 684          */
 685         if (frame->tf_err & PGEX_RSV) {
 686                 trap_fatal(frame, eva);
 687                 return (-1);
 688         }
 689
 690         /*
 691          * PGEX_I is defined only if the execute disable bit capability is
 692          * supported and enabled.
 693          */
 694         if (frame->tf_err & PGEX_W)
 695                 ftype = VM_PROT_WRITE;
 696         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 697                 ftype = VM_PROT_EXECUTE;
 698         else
 699                 ftype = VM_PROT_READ;
 700
 701         /* Fault in the page. */
 702         rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 703         if (rv == KERN_SUCCESS) {
 704 #ifdef HWPMC_HOOKS
 705                 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
 706                         PMC_SOFT_CALL_TF( , , page_fault, all, frame);
 707                         if (ftype == VM_PROT_READ)
 708                                 PMC_SOFT_CALL_TF( , , page_fault, read,
 709                                     frame);
 710                         else
 711                                 PMC_SOFT_CALL_TF( , , page_fault, write,
 712                                     frame);
 713                 }
 714 #endif
 715                 return (0);
 716         }
 717         if (!usermode) {
 718                 if (td->td_intr_nesting_level == 0 &&
 719                     curpcb->pcb_onfault != NULL) {
 720                         frame->tf_rip = (long)curpcb->pcb_onfault;
 721                         return (0);
 722                 }
 723                 trap_fatal(frame, eva);
 724                 return (-1);
 725         }
 726         return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 727 }
 728
 729 static void
 730 trap_fatal(frame, eva)
 731         struct trapframe *frame;
 732         vm_offset_t eva;
 733 {
 734         int code, ss;
 735         u_int type;
 736         struct soft_segment_descriptor softseg;
 737         char *msg;
 738
 739         code = frame->tf_err;
 740         type = frame->tf_trapno;
 741         sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)],
 742             &softseg);
 743
 744         if (type <= MAX_TRAP_MSG)
 745                 msg = trap_msg[type];
 746         else
 747                 msg = "UNKNOWN";
 748         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 749             TRAPF_USERMODE(frame) ? "user" : "kernel");
 750 #ifdef SMP
 751         /* two separate prints in case of a trap on an unmapped page */
 752         printf("cpuid = %d; ", PCPU_GET(cpuid));
 753         printf("apic id = %02x\n", PCPU_GET(apic_id));
 754 #endif
 755         if (type == T_PAGEFLT) {
 756                 printf("fault virtual address   = 0x%lx\n", eva);
 757                 printf("fault code              = %s %s %s, %s\n",
 758                         code & PGEX_U ? "user" : "supervisor",
 759                         code & PGEX_W ? "write" : "read",
 760                         code & PGEX_I ? "instruction" : "data",
 761                         code & PGEX_RSV ? "reserved bits in PTE" :
 762                         code & PGEX_P ? "protection violation" : "page not present");
 763         }
 764         printf("instruction pointer     = 0x%lx:0x%lx\n",
 765                frame->tf_cs & 0xffff, frame->tf_rip);
 766         ss = frame->tf_ss & 0xffff;
 767         printf("stack pointer           = 0x%x:0x%lx\n", ss, frame->tf_rsp);
 768         printf("frame pointer           = 0x%x:0x%lx\n", ss, frame->tf_rbp);
 769         printf("code segment            = base 0x%lx, limit 0x%lx, type 0x%x\n",
 770                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 771         printf("                        = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
 772                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
 773                softseg.ssd_gran);
 774         printf("processor eflags        = ");
 775         if (frame->tf_rflags & PSL_T)
 776                 printf("trace trap, ");
 777         if (frame->tf_rflags & PSL_I)
 778                 printf("interrupt enabled, ");
 779         if (frame->tf_rflags & PSL_NT)
 780                 printf("nested task, ");
 781         if (frame->tf_rflags & PSL_RF)
 782                 printf("resume, ");
 783         printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
 784         printf("current process         = %d (%s)\n",
 785             curproc->p_pid, curthread->td_name);
 786
 787 #ifdef KDB
 788         if (debugger_on_panic || kdb_active)
 789                 if (kdb_trap(type, 0, frame))
 790                         return;
 791 #endif
 792         printf("trap number             = %d\n", type);
 793         if (type <= MAX_TRAP_MSG)
 794                 panic("%s", trap_msg[type]);
 795         else
 796                 panic("unknown/reserved trap");
 797 }
 798
 799 /*
 800  * Double fault handler. Called when a fault occurs while writing
 801  * a frame for a trap/exception onto the stack. This usually occurs
 802  * when the stack overflows (such is the case with infinite recursion,
 803  * for example).
 804  */
 805 void
 806 dblfault_handler(struct trapframe *frame)
 807 {
 808 #ifdef KDTRACE_HOOKS
 809         if (dtrace_doubletrap_func != NULL)
 810                 (*dtrace_doubletrap_func)();
 811 #endif
 812         printf("\nFatal double fault\n"
 813             "rip %#lx rsp %#lx rbp %#lx\n"
 814             "rax %#lx rdx %#lx rbx %#lx\n"
 815             "rcx %#lx rsi %#lx rdi %#lx\n"
 816             "r8 %#lx r9 %#lx r10 %#lx\n"
 817             "r11 %#lx r12 %#lx r13 %#lx\n"
 818             "r14 %#lx r15 %#lx rflags %#lx\n"
 819             "cs %#lx ss %#lx ds %#hx es %#hx fs %#hx gs %#hx\n"
 820             "fsbase %#lx gsbase %#lx kgsbase %#lx\n",
 821             frame->tf_rip, frame->tf_rsp, frame->tf_rbp,
 822             frame->tf_rax, frame->tf_rdx, frame->tf_rbx,
 823             frame->tf_rcx, frame->tf_rdi, frame->tf_rsi,
 824             frame->tf_r8, frame->tf_r9, frame->tf_r10,
 825             frame->tf_r11, frame->tf_r12, frame->tf_r13,
 826             frame->tf_r14, frame->tf_r15, frame->tf_rflags,
 827             frame->tf_cs, frame->tf_ss, frame->tf_ds, frame->tf_es,
 828             frame->tf_fs, frame->tf_gs,
 829             rdmsr(MSR_FSBASE), rdmsr(MSR_GSBASE), rdmsr(MSR_KGSBASE));
 830 #ifdef SMP
 831         /* two separate prints in case of a trap on an unmapped page */
 832         printf("cpuid = %d; ", PCPU_GET(cpuid));
 833         printf("apic id = %02x\n", PCPU_GET(apic_id));
 834 #endif
 835         panic("double fault");
 836 }
 837
 838 int
 839 cpu_fetch_syscall_args(struct thread *td)
 840 {
 841         struct proc *p;
 842         struct trapframe *frame;
 843         register_t *argp;
 844         struct syscall_args *sa;
 845         caddr_t params;
 846         int reg, regcnt, error;
 847
 848         p = td->td_proc;
 849         frame = td->td_frame;
 850         sa = &td->td_sa;
 851         reg = 0;
 852         regcnt = 6;
 853
 854         params = (caddr_t)frame->tf_rsp + sizeof(register_t);
 855         sa->code = frame->tf_rax;
 856
 857         if (sa->code == SYS_syscall || sa->code == SYS___syscall) {
 858                 sa->code = frame->tf_rdi;
 859                 reg++;
 860                 regcnt--;
 861         }
 862         if (p->p_sysent->sv_mask)
 863                 sa->code &= p->p_sysent->sv_mask;
 864
 865         if (sa->code >= p->p_sysent->sv_size)
 866                 sa->callp = &p->p_sysent->sv_table[0];
 867         else
 868                 sa->callp = &p->p_sysent->sv_table[sa->code];
 869
 870         sa->narg = sa->callp->sy_narg;
 871         KASSERT(sa->narg <= sizeof(sa->args) / sizeof(sa->args[0]),
 872             ("Too many syscall arguments!"));
 873         error = 0;
 874         argp = &frame->tf_rdi;
 875         argp += reg;
 876         bcopy(argp, sa->args, sizeof(sa->args[0]) * regcnt);
 877         if (sa->narg > regcnt) {
 878                 KASSERT(params != NULL, ("copyin args with no params!"));
 879                 error = copyin(params, &sa->args[regcnt],
 880                     (sa->narg - regcnt) * sizeof(sa->args[0]));
 881         }
 882
 883         if (error == 0) {
 884                 td->td_retval[0] = 0;
 885                 td->td_retval[1] = frame->tf_rdx;
 886         }
 887
 888         return (error);
 889 }
 890
 891 #include "../../kern/subr_syscall.c"
 892
 893 /*
 894  * System call handler for native binaries.  The trap frame is already
 895  * set up by the assembler trampoline and a pointer to it is saved in
 896  * td_frame.
 897  */
 898 void
 899 amd64_syscall(struct thread *td, int traced)
 900 {
 901         int error;
 902         ksiginfo_t ksi;
 903
 904 #ifdef DIAGNOSTIC
 905         if (!TRAPF_USERMODE(td->td_frame)) {
 906                 panic("syscall");
 907                 /* NOT REACHED */
 908         }
 909 #endif
 910         error = syscallenter(td);
 911
 912         /*
 913          * Traced syscall.
 914          */
 915         if (__predict_false(traced)) {
 916                 td->td_frame->tf_rflags &= ~PSL_T;
 917                 ksiginfo_init_trap(&ksi);
 918                 ksi.ksi_signo = SIGTRAP;
 919                 ksi.ksi_code = TRAP_TRACE;
 920                 ksi.ksi_addr = (void *)td->td_frame->tf_rip;
 921                 trapsignal(td, &ksi);
 922         }
 923
 924         KASSERT(PCB_USER_FPU(td->td_pcb),
 925             ("System call %s returning with kernel FPU ctx leaked",
 926              syscallname(td->td_proc, td->td_sa.code)));
 927         KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td),
 928             ("System call %s returning with mangled pcb_save",
 929              syscallname(td->td_proc, td->td_sa.code)));
 930         KASSERT(td->td_md.md_invl_gen.gen == 0,
 931             ("System call %s returning with leaked invl_gen %lu",
 932             syscallname(td->td_proc, td->td_sa.code),
 933             td->td_md.md_invl_gen.gen));
 934
 935         syscallret(td, error);
 936
 937         /*
 938          * If the user-supplied value of %rip is not a canonical
 939          * address, then some CPUs will trigger a ring 0 #GP during
 940          * the sysret instruction.  However, the fault handler would
 941          * execute in ring 0 with the user's %gs and %rsp which would
 942          * not be safe.  Instead, use the full return path which
 943          * catches the problem safely.
 944          */
 945         if (__predict_false(td->td_frame->tf_rip >= VM_MAXUSER_ADDRESS))
 946                 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
 947 }