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 void __noinline trap(struct trapframe *frame);
 103 void trap_check(struct trapframe *frame);
 104 void dblfault_handler(struct trapframe *frame);
 105
 106 static int trap_pfault(struct trapframe *, int);
 107 static void trap_fatal(struct trapframe *, vm_offset_t);
 108
 109 #define MAX_TRAP_MSG            32
 110 static char *trap_msg[] = {
 111         "",                                     /*  0 unused */
 112         "privileged instruction fault",         /*  1 T_PRIVINFLT */
 113         "",                                     /*  2 unused */
 114         "breakpoint instruction fault",         /*  3 T_BPTFLT */
 115         "",                                     /*  4 unused */
 116         "",                                     /*  5 unused */
 117         "arithmetic trap",                      /*  6 T_ARITHTRAP */
 118         "",                                     /*  7 unused */
 119         "",                                     /*  8 unused */
 120         "general protection fault",             /*  9 T_PROTFLT */
 121         "trace trap",                           /* 10 T_TRCTRAP */
 122         "",                                     /* 11 unused */
 123         "page fault",                           /* 12 T_PAGEFLT */
 124         "",                                     /* 13 unused */
 125         "alignment fault",                      /* 14 T_ALIGNFLT */
 126         "",                                     /* 15 unused */
 127         "",                                     /* 16 unused */
 128         "",                                     /* 17 unused */
 129         "integer divide fault",                 /* 18 T_DIVIDE */
 130         "non-maskable interrupt trap",          /* 19 T_NMI */
 131         "overflow trap",                        /* 20 T_OFLOW */
 132         "FPU bounds check fault",               /* 21 T_BOUND */
 133         "FPU device not available",             /* 22 T_DNA */
 134         "double fault",                         /* 23 T_DOUBLEFLT */
 135         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
 136         "invalid TSS fault",                    /* 25 T_TSSFLT */
 137         "segment not present fault",            /* 26 T_SEGNPFLT */
 138         "stack fault",                          /* 27 T_STKFLT */
 139         "machine check trap",                   /* 28 T_MCHK */
 140         "SIMD floating-point exception",        /* 29 T_XMMFLT */
 141         "reserved (unknown) fault",             /* 30 T_RESERVED */
 142         "",                                     /* 31 unused (reserved) */
 143         "DTrace pid return trap",               /* 32 T_DTRACE_RET */
 144 };
 145
 146 static int prot_fault_translation;
 147 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RWTUN,
 148     &prot_fault_translation, 0,
 149     "Select signal to deliver on protection fault");
 150 static int uprintf_signal;
 151 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RWTUN,
 152     &uprintf_signal, 0,
 153     "Print debugging information on trap signal to ctty");
 154
 155 /*
 156  * Exception, fault, and trap interface to the FreeBSD kernel.
 157  * This common code is called from assembly language IDT gate entry
 158  * routines that prepare a suitable stack frame, and restore this
 159  * frame after the exception has been processed.
 160  */
 161
 162 void
 163 trap(struct trapframe *frame)
 164 {
 165 #ifdef KDTRACE_HOOKS
 166         struct reg regs;
 167 #endif
 168         ksiginfo_t ksi;
 169         struct thread *td;
 170         struct proc *p;
 171         register_t addr;
 172 #ifdef KDB
 173         register_t dr6;
 174 #endif
 175         int signo, ucode;
 176         u_int type;
 177
 178         td = curthread;
 179         p = td->td_proc;
 180         signo = 0;
 181         ucode = 0;
 182         addr = 0;
 183
 184         VM_CNT_INC(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 && ipi_nmi_handler() == 0)
 190                 return;
 191 #endif
 192
 193 #ifdef KDB
 194         if (kdb_active) {
 195                 kdb_reenter();
 196                 return;
 197         }
 198 #endif
 199
 200         if (type == T_RESERVED) {
 201                 trap_fatal(frame, 0);
 202                 return;
 203         }
 204
 205         if (type == T_NMI) {
 206 #ifdef HWPMC_HOOKS
 207                 /*
 208                  * CPU PMCs interrupt using an NMI.  If the PMC module is
 209                  * active, pass the 'rip' value to the PMC module's interrupt
 210                  * handler.  A non-zero return value from the handler means that
 211                  * the NMI was consumed by it and we can return immediately.
 212                  */
 213                 if (pmc_intr != NULL &&
 214                     (*pmc_intr)(PCPU_GET(cpuid), frame) != 0)
 215                         return;
 216 #endif
 217
 218 #ifdef STACK
 219                 if (stack_nmi_handler(frame) != 0)
 220                         return;
 221 #endif
 222         }
 223
 224         if (type == T_MCHK) {
 225                 mca_intr();
 226                 return;
 227         }
 228
 229         if ((frame->tf_rflags & PSL_I) == 0) {
 230                 /*
 231                  * Buggy application or kernel code has disabled
 232                  * interrupts and then trapped.  Enabling interrupts
 233                  * now is wrong, but it is better than running with
 234                  * interrupts disabled until they are accidentally
 235                  * enabled later.
 236                  */
 237                 if (TRAPF_USERMODE(frame))
 238                         uprintf(
 239                             "pid %ld (%s): trap %d with interrupts disabled\n",
 240                             (long)curproc->p_pid, curthread->td_name, type);
 241                 else if (type != T_NMI && type != T_BPTFLT &&
 242                     type != T_TRCTRAP) {
 243                         /*
 244                          * XXX not quite right, since this may be for a
 245                          * multiple fault in user mode.
 246                          */
 247                         printf("kernel trap %d with interrupts disabled\n",
 248                             type);
 249
 250                         /*
 251                          * We shouldn't enable interrupts while holding a
 252                          * spin lock.
 253                          */
 254                         if (td->td_md.md_spinlock_count == 0)
 255                                 enable_intr();
 256                 }
 257         }
 258
 259         if (TRAPF_USERMODE(frame)) {
 260                 /* user trap */
 261
 262                 td->td_pticks = 0;
 263                 td->td_frame = frame;
 264                 addr = frame->tf_rip;
 265                 if (td->td_cowgen != p->p_cowgen)
 266                         thread_cow_update(td);
 267
 268                 switch (type) {
 269                 case T_PRIVINFLT:       /* privileged instruction fault */
 270                         signo = SIGILL;
 271                         ucode = ILL_PRVOPC;
 272                         break;
 273
 274                 case T_BPTFLT:          /* bpt instruction fault */
 275                 case T_TRCTRAP:         /* trace trap */
 276                         enable_intr();
 277 #ifdef KDTRACE_HOOKS
 278                         if (type == T_BPTFLT) {
 279                                 fill_frame_regs(frame, &regs);
 280                                 if (dtrace_pid_probe_ptr != NULL &&
 281                                     dtrace_pid_probe_ptr(&regs) == 0)
 282                                         return;
 283                         }
 284 #endif
 285                         frame->tf_rflags &= ~PSL_T;
 286                         signo = SIGTRAP;
 287                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 288                         break;
 289
 290                 case T_ARITHTRAP:       /* arithmetic trap */
 291                         ucode = fputrap_x87();
 292                         if (ucode == -1)
 293                                 return;
 294                         signo = SIGFPE;
 295                         break;
 296
 297                 case T_PROTFLT:         /* general protection fault */
 298                         signo = SIGBUS;
 299                         ucode = BUS_OBJERR;
 300                         break;
 301                 case T_STKFLT:          /* stack fault */
 302                 case T_SEGNPFLT:        /* segment not present fault */
 303                         signo = SIGBUS;
 304                         ucode = BUS_ADRERR;
 305                         break;
 306                 case T_TSSFLT:          /* invalid TSS fault */
 307                         signo = SIGBUS;
 308                         ucode = BUS_OBJERR;
 309                         break;
 310                 case T_ALIGNFLT:
 311                         signo = SIGBUS;
 312                         ucode = BUS_ADRALN;
 313                         break;
 314                 case T_DOUBLEFLT:       /* double fault */
 315                 default:
 316                         signo = SIGBUS;
 317                         ucode = BUS_OBJERR;
 318                         break;
 319
 320                 case T_PAGEFLT:         /* page fault */
 321                         /*
 322                          * Emulator can take care about this trap?
 323                          */
 324                         if (*p->p_sysent->sv_trap != NULL &&
 325                             (*p->p_sysent->sv_trap)(td) == 0)
 326                                 return;
 327
 328                         addr = frame->tf_addr;
 329                         signo = trap_pfault(frame, TRUE);
 330                         if (signo == -1)
 331                                 return;
 332                         if (signo == 0)
 333                                 goto userret;
 334                         if (signo == SIGSEGV) {
 335                                 ucode = SEGV_MAPERR;
 336                         } else if (prot_fault_translation == 0) {
 337                                 /*
 338                                  * Autodetect.  This check also covers
 339                                  * the images without the ABI-tag ELF
 340                                  * note.
 341                                  */
 342                                 if (SV_CURPROC_ABI() == SV_ABI_FREEBSD &&
 343                                     p->p_osrel >= P_OSREL_SIGSEGV) {
 344                                         signo = SIGSEGV;
 345                                         ucode = SEGV_ACCERR;
 346                                 } else {
 347                                         signo = SIGBUS;
 348                                         ucode = BUS_PAGE_FAULT;
 349                                 }
 350                         } else if (prot_fault_translation == 1) {
 351                                 /*
 352                                  * Always compat mode.
 353                                  */
 354                                 signo = SIGBUS;
 355                                 ucode = BUS_PAGE_FAULT;
 356                         } else {
 357                                 /*
 358                                  * Always SIGSEGV mode.
 359                                  */
 360                                 signo = SIGSEGV;
 361                                 ucode = SEGV_ACCERR;
 362                         }
 363                         break;
 364
 365                 case T_DIVIDE:          /* integer divide fault */
 366                         ucode = FPE_INTDIV;
 367                         signo = SIGFPE;
 368                         break;
 369
 370 #ifdef DEV_ISA
 371                 case T_NMI:
 372                         nmi_handle_intr(type, frame);
 373                         return;
 374 #endif
 375
 376                 case T_OFLOW:           /* integer overflow fault */
 377                         ucode = FPE_INTOVF;
 378                         signo = SIGFPE;
 379                         break;
 380
 381                 case T_BOUND:           /* bounds check fault */
 382                         ucode = FPE_FLTSUB;
 383                         signo = SIGFPE;
 384                         break;
 385
 386                 case T_DNA:
 387                         /* transparent fault (due to context switch "late") */
 388                         KASSERT(PCB_USER_FPU(td->td_pcb),
 389                             ("kernel FPU ctx has leaked"));
 390                         fpudna();
 391                         return;
 392
 393                 case T_FPOPFLT:         /* FPU operand fetch fault */
 394                         ucode = ILL_COPROC;
 395                         signo = SIGILL;
 396                         break;
 397
 398                 case T_XMMFLT:          /* SIMD floating-point exception */
 399                         ucode = fputrap_sse();
 400                         if (ucode == -1)
 401                                 return;
 402                         signo = SIGFPE;
 403                         break;
 404 #ifdef KDTRACE_HOOKS
 405                 case T_DTRACE_RET:
 406                         enable_intr();
 407                         fill_frame_regs(frame, &regs);
 408                         if (dtrace_return_probe_ptr != NULL)
 409                                 dtrace_return_probe_ptr(&regs);
 410                         return;
 411 #endif
 412                 }
 413         } else {
 414                 /* kernel trap */
 415
 416                 KASSERT(cold || td->td_ucred != NULL,
 417                     ("kernel trap doesn't have ucred"));
 418                 switch (type) {
 419                 case T_PAGEFLT:                 /* page fault */
 420                         (void) trap_pfault(frame, FALSE);
 421                         return;
 422
 423                 case T_DNA:
 424                         if (PCB_USER_FPU(td->td_pcb))
 425                                 panic("Unregistered use of FPU in kernel");
 426                         fpudna();
 427                         return;
 428
 429                 case T_ARITHTRAP:       /* arithmetic trap */
 430                 case T_XMMFLT:          /* SIMD floating-point exception */
 431                 case T_FPOPFLT:         /* FPU operand fetch fault */
 432                         /*
 433                          * For now, supporting kernel handler
 434                          * registration for FPU traps is overkill.
 435                          */
 436                         trap_fatal(frame, 0);
 437                         return;
 438
 439                 case T_STKFLT:          /* stack fault */
 440                 case T_PROTFLT:         /* general protection fault */
 441                 case T_SEGNPFLT:        /* segment not present fault */
 442                         if (td->td_intr_nesting_level != 0)
 443                                 break;
 444
 445                         /*
 446                          * Invalid segment selectors and out of bounds
 447                          * %rip's and %rsp's can be set up in user mode.
 448                          * This causes a fault in kernel mode when the
 449                          * kernel tries to return to user mode.  We want
 450                          * to get this fault so that we can fix the
 451                          * problem here and not have to check all the
 452                          * selectors and pointers when the user changes
 453                          * them.
 454                          */
 455                         if (frame->tf_rip == (long)doreti_iret) {
 456                                 frame->tf_rip = (long)doreti_iret_fault;
 457                                 return;
 458                         }
 459                         if (frame->tf_rip == (long)ld_ds) {
 460                                 frame->tf_rip = (long)ds_load_fault;
 461                                 return;
 462                         }
 463                         if (frame->tf_rip == (long)ld_es) {
 464                                 frame->tf_rip = (long)es_load_fault;
 465                                 return;
 466                         }
 467                         if (frame->tf_rip == (long)ld_fs) {
 468                                 frame->tf_rip = (long)fs_load_fault;
 469                                 return;
 470                         }
 471                         if (frame->tf_rip == (long)ld_gs) {
 472                                 frame->tf_rip = (long)gs_load_fault;
 473                                 return;
 474                         }
 475                         if (frame->tf_rip == (long)ld_gsbase) {
 476                                 frame->tf_rip = (long)gsbase_load_fault;
 477                                 return;
 478                         }
 479                         if (frame->tf_rip == (long)ld_fsbase) {
 480                                 frame->tf_rip = (long)fsbase_load_fault;
 481                                 return;
 482                         }
 483                         if (curpcb->pcb_onfault != NULL) {
 484                                 frame->tf_rip = (long)curpcb->pcb_onfault;
 485                                 return;
 486                         }
 487                         break;
 488
 489                 case T_TSSFLT:
 490                         /*
 491                          * PSL_NT can be set in user mode and isn't cleared
 492                          * automatically when the kernel is entered.  This
 493                          * causes a TSS fault when the kernel attempts to
 494                          * `iret' because the TSS link is uninitialized.  We
 495                          * want to get this fault so that we can fix the
 496                          * problem here and not every time the kernel is
 497                          * entered.
 498                          */
 499                         if (frame->tf_rflags & PSL_NT) {
 500                                 frame->tf_rflags &= ~PSL_NT;
 501                                 return;
 502                         }
 503                         break;
 504
 505                 case T_TRCTRAP:  /* trace trap */
 506                         /*
 507                          * Ignore debug register trace traps due to
 508                          * accesses in the user's address space, which
 509                          * can happen under several conditions such as
 510                          * if a user sets a watchpoint on a buffer and
 511                          * then passes that buffer to a system call.
 512                          * We still want to get TRCTRAPS for addresses
 513                          * in kernel space because that is useful when
 514                          * debugging the kernel.
 515                          */
 516                         if (user_dbreg_trap()) {
 517                                 /*
 518                                  * Reset breakpoint bits because the
 519                                  * processor doesn't
 520                                  */
 521                                 load_dr6(rdr6() & ~0xf);
 522                                 return;
 523                         }
 524                         /*
 525                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 526                          */
 527                 case T_BPTFLT:
 528                         /*
 529                          * If KDB is enabled, let it handle the debugger trap.
 530                          * Otherwise, debugger traps "can't happen".
 531                          */
 532 #ifdef KDB
 533                         /* XXX %dr6 is not quite reentrant. */
 534                         dr6 = rdr6();
 535                         load_dr6(dr6 & ~0x4000);
 536                         if (kdb_trap(type, dr6, frame))
 537                                 return;
 538 #endif
 539                         break;
 540
 541 #ifdef DEV_ISA
 542                 case T_NMI:
 543                         nmi_handle_intr(type, frame);
 544                         return;
 545 #endif
 546                 }
 547
 548                 trap_fatal(frame, 0);
 549                 return;
 550         }
 551
 552         /* Translate fault for emulators (e.g. Linux) */
 553         if (*p->p_sysent->sv_transtrap != NULL)
 554                 signo = (*p->p_sysent->sv_transtrap)(signo, type);
 555
 556         ksiginfo_init_trap(&ksi);
 557         ksi.ksi_signo = signo;
 558         ksi.ksi_code = ucode;
 559         ksi.ksi_trapno = type;
 560         ksi.ksi_addr = (void *)addr;
 561         if (uprintf_signal) {
 562                 uprintf("pid %d comm %s: signal %d err %lx code %d type %d "
 563                     "addr 0x%lx rsp 0x%lx rip 0x%lx "
 564                     "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
 565                     p->p_pid, p->p_comm, signo, frame->tf_err, ucode, type,
 566                     addr, frame->tf_rsp, frame->tf_rip,
 567                     fubyte((void *)(frame->tf_rip + 0)),
 568                     fubyte((void *)(frame->tf_rip + 1)),
 569                     fubyte((void *)(frame->tf_rip + 2)),
 570                     fubyte((void *)(frame->tf_rip + 3)),
 571                     fubyte((void *)(frame->tf_rip + 4)),
 572                     fubyte((void *)(frame->tf_rip + 5)),
 573                     fubyte((void *)(frame->tf_rip + 6)),
 574                     fubyte((void *)(frame->tf_rip + 7)));
 575         }
 576         KASSERT((read_rflags() & PSL_I) != 0, ("interrupts disabled"));
 577         trapsignal(td, &ksi);
 578 userret:
 579         userret(td, frame);
 580         KASSERT(PCB_USER_FPU(td->td_pcb),
 581             ("Return from trap with kernel FPU ctx leaked"));
 582 }
 583
 584 /*
 585  * Ensure that we ignore any DTrace-induced faults. This function cannot
 586  * be instrumented, so it cannot generate such faults itself.
 587  */
 588 void
 589 trap_check(struct trapframe *frame)
 590 {
 591
 592 #ifdef KDTRACE_HOOKS
 593         if (dtrace_trap_func != NULL &&
 594             (*dtrace_trap_func)(frame, frame->tf_trapno) != 0)
 595                 return;
 596 #endif
 597         trap(frame);
 598 }
 599
 600 static int
 601 trap_pfault(struct trapframe *frame, int usermode)
 602 {
 603         struct thread *td;
 604         struct proc *p;
 605         vm_map_t map;
 606         vm_offset_t va;
 607         int rv;
 608         vm_prot_t ftype;
 609         vm_offset_t eva;
 610
 611         td = curthread;
 612         p = td->td_proc;
 613         eva = frame->tf_addr;
 614         rv = 0;
 615
 616         if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
 617                 /*
 618                  * Due to both processor errata and lazy TLB invalidation when
 619                  * access restrictions are removed from virtual pages, memory
 620                  * accesses that are allowed by the physical mapping layer may
 621                  * nonetheless cause one spurious page fault per virtual page.
 622                  * When the thread is executing a "no faulting" section that
 623                  * is bracketed by vm_fault_{disable,enable}_pagefaults(),
 624                  * every page fault is treated as a spurious page fault,
 625                  * unless it accesses the same virtual address as the most
 626                  * recent page fault within the same "no faulting" section.
 627                  */
 628                 if (td->td_md.md_spurflt_addr != eva ||
 629                     (td->td_pflags & TDP_RESETSPUR) != 0) {
 630                         /*
 631                          * Do nothing to the TLB.  A stale TLB entry is
 632                          * flushed automatically by a page fault.
 633                          */
 634                         td->td_md.md_spurflt_addr = eva;
 635                         td->td_pflags &= ~TDP_RESETSPUR;
 636                         return (0);
 637                 }
 638         } else {
 639                 /*
 640                  * If we get a page fault while in a critical section, then
 641                  * it is most likely a fatal kernel page fault.  The kernel
 642                  * is already going to panic trying to get a sleep lock to
 643                  * do the VM lookup, so just consider it a fatal trap so the
 644                  * kernel can print out a useful trap message and even get
 645                  * to the debugger.
 646                  *
 647                  * If we get a page fault while holding a non-sleepable
 648                  * lock, then it is most likely a fatal kernel page fault.
 649                  * If WITNESS is enabled, then it's going to whine about
 650                  * bogus LORs with various VM locks, so just skip to the
 651                  * fatal trap handling directly.
 652                  */
 653                 if (td->td_critnest != 0 ||
 654                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 655                     "Kernel page fault") != 0) {
 656                         trap_fatal(frame, eva);
 657                         return (-1);
 658                 }
 659         }
 660         va = trunc_page(eva);
 661         if (va >= VM_MIN_KERNEL_ADDRESS) {
 662                 /*
 663                  * Don't allow user-mode faults in kernel address space.
 664                  */
 665                 if (usermode)
 666                         goto nogo;
 667
 668                 map = kernel_map;
 669         } else {
 670                 map = &p->p_vmspace->vm_map;
 671
 672                 /*
 673                  * When accessing a usermode address, kernel must be
 674                  * ready to accept the page fault, and provide a
 675                  * handling routine.  Since accessing the address
 676                  * without the handler is a bug, do not try to handle
 677                  * it normally, and panic immediately.
 678                  */
 679                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 680                     curpcb->pcb_onfault == NULL)) {
 681                         trap_fatal(frame, eva);
 682                         return (-1);
 683                 }
 684         }
 685
 686         /*
 687          * If the trap was caused by errant bits in the PTE then panic.
 688          */
 689         if (frame->tf_err & PGEX_RSV) {
 690                 trap_fatal(frame, eva);
 691                 return (-1);
 692         }
 693
 694         /*
 695          * PGEX_I is defined only if the execute disable bit capability is
 696          * supported and enabled.
 697          */
 698         if (frame->tf_err & PGEX_W)
 699                 ftype = VM_PROT_WRITE;
 700         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 701                 ftype = VM_PROT_EXECUTE;
 702         else
 703                 ftype = VM_PROT_READ;
 704
 705         /* Fault in the page. */
 706         rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 707         if (rv == KERN_SUCCESS) {
 708 #ifdef HWPMC_HOOKS
 709                 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
 710                         PMC_SOFT_CALL_TF( , , page_fault, all, frame);
 711                         if (ftype == VM_PROT_READ)
 712                                 PMC_SOFT_CALL_TF( , , page_fault, read,
 713                                     frame);
 714                         else
 715                                 PMC_SOFT_CALL_TF( , , page_fault, write,
 716                                     frame);
 717                 }
 718 #endif
 719                 return (0);
 720         }
 721 nogo:
 722         if (!usermode) {
 723                 if (td->td_intr_nesting_level == 0 &&
 724                     curpcb->pcb_onfault != NULL) {
 725                         frame->tf_rip = (long)curpcb->pcb_onfault;
 726                         return (0);
 727                 }
 728                 trap_fatal(frame, eva);
 729                 return (-1);
 730         }
 731         return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 732 }
 733
 734 static void
 735 trap_fatal(frame, eva)
 736         struct trapframe *frame;
 737         vm_offset_t eva;
 738 {
 739         int code, ss;
 740         u_int type;
 741         struct soft_segment_descriptor softseg;
 742         char *msg;
 743
 744         code = frame->tf_err;
 745         type = frame->tf_trapno;
 746         sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)],
 747             &softseg);
 748
 749         if (type <= MAX_TRAP_MSG)
 750                 msg = trap_msg[type];
 751         else
 752                 msg = "UNKNOWN";
 753         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 754             TRAPF_USERMODE(frame) ? "user" : "kernel");
 755 #ifdef SMP
 756         /* two separate prints in case of a trap on an unmapped page */
 757         printf("cpuid = %d; ", PCPU_GET(cpuid));
 758         printf("apic id = %02x\n", PCPU_GET(apic_id));
 759 #endif
 760         if (type == T_PAGEFLT) {
 761                 printf("fault virtual address   = 0x%lx\n", eva);
 762                 printf("fault code              = %s %s %s, %s\n",
 763                         code & PGEX_U ? "user" : "supervisor",
 764                         code & PGEX_W ? "write" : "read",
 765                         code & PGEX_I ? "instruction" : "data",
 766                         code & PGEX_RSV ? "reserved bits in PTE" :
 767                         code & PGEX_P ? "protection violation" : "page not present");
 768         }
 769         printf("instruction pointer     = 0x%lx:0x%lx\n",
 770                frame->tf_cs & 0xffff, frame->tf_rip);
 771         ss = frame->tf_ss & 0xffff;
 772         printf("stack pointer           = 0x%x:0x%lx\n", ss, frame->tf_rsp);
 773         printf("frame pointer           = 0x%x:0x%lx\n", ss, frame->tf_rbp);
 774         printf("code segment            = base 0x%lx, limit 0x%lx, type 0x%x\n",
 775                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 776         printf("                        = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
 777                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
 778                softseg.ssd_gran);
 779         printf("processor eflags        = ");
 780         if (frame->tf_rflags & PSL_T)
 781                 printf("trace trap, ");
 782         if (frame->tf_rflags & PSL_I)
 783                 printf("interrupt enabled, ");
 784         if (frame->tf_rflags & PSL_NT)
 785                 printf("nested task, ");
 786         if (frame->tf_rflags & PSL_RF)
 787                 printf("resume, ");
 788         printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
 789         printf("current process         = %d (%s)\n",
 790             curproc->p_pid, curthread->td_name);
 791
 792 #ifdef KDB
 793         if (debugger_on_panic || kdb_active)
 794                 if (kdb_trap(type, 0, frame))
 795                         return;
 796 #endif
 797         printf("trap number             = %d\n", type);
 798         if (type <= MAX_TRAP_MSG)
 799                 panic("%s", trap_msg[type]);
 800         else
 801                 panic("unknown/reserved trap");
 802 }
 803
 804 /*
 805  * Double fault handler. Called when a fault occurs while writing
 806  * a frame for a trap/exception onto the stack. This usually occurs
 807  * when the stack overflows (such is the case with infinite recursion,
 808  * for example).
 809  */
 810 void
 811 dblfault_handler(struct trapframe *frame)
 812 {
 813 #ifdef KDTRACE_HOOKS
 814         if (dtrace_doubletrap_func != NULL)
 815                 (*dtrace_doubletrap_func)();
 816 #endif
 817         printf("\nFatal double fault\n"
 818             "rip %#lx rsp %#lx rbp %#lx\n"
 819             "rax %#lx rdx %#lx rbx %#lx\n"
 820             "rcx %#lx rsi %#lx rdi %#lx\n"
 821             "r8 %#lx r9 %#lx r10 %#lx\n"
 822             "r11 %#lx r12 %#lx r13 %#lx\n"
 823             "r14 %#lx r15 %#lx rflags %#lx\n"
 824             "cs %#lx ss %#lx ds %#hx es %#hx fs %#hx gs %#hx\n"
 825             "fsbase %#lx gsbase %#lx kgsbase %#lx\n",
 826             frame->tf_rip, frame->tf_rsp, frame->tf_rbp,
 827             frame->tf_rax, frame->tf_rdx, frame->tf_rbx,
 828             frame->tf_rcx, frame->tf_rdi, frame->tf_rsi,
 829             frame->tf_r8, frame->tf_r9, frame->tf_r10,
 830             frame->tf_r11, frame->tf_r12, frame->tf_r13,
 831             frame->tf_r14, frame->tf_r15, frame->tf_rflags,
 832             frame->tf_cs, frame->tf_ss, frame->tf_ds, frame->tf_es,
 833             frame->tf_fs, frame->tf_gs,
 834             rdmsr(MSR_FSBASE), rdmsr(MSR_GSBASE), rdmsr(MSR_KGSBASE));
 835 #ifdef SMP
 836         /* two separate prints in case of a trap on an unmapped page */
 837         printf("cpuid = %d; ", PCPU_GET(cpuid));
 838         printf("apic id = %02x\n", PCPU_GET(apic_id));
 839 #endif
 840         panic("double fault");
 841 }
 842
 843 int
 844 cpu_fetch_syscall_args(struct thread *td)
 845 {
 846         struct proc *p;
 847         struct trapframe *frame;
 848         register_t *argp;
 849         struct syscall_args *sa;
 850         caddr_t params;
 851         int reg, regcnt, error;
 852
 853         p = td->td_proc;
 854         frame = td->td_frame;
 855         sa = &td->td_sa;
 856         reg = 0;
 857         regcnt = 6;
 858
 859         params = (caddr_t)frame->tf_rsp + sizeof(register_t);
 860         sa->code = frame->tf_rax;
 861
 862         if (sa->code == SYS_syscall || sa->code == SYS___syscall) {
 863                 sa->code = frame->tf_rdi;
 864                 reg++;
 865                 regcnt--;
 866         }
 867         if (p->p_sysent->sv_mask)
 868                 sa->code &= p->p_sysent->sv_mask;
 869
 870         if (sa->code >= p->p_sysent->sv_size)
 871                 sa->callp = &p->p_sysent->sv_table[0];
 872         else
 873                 sa->callp = &p->p_sysent->sv_table[sa->code];
 874
 875         sa->narg = sa->callp->sy_narg;
 876         KASSERT(sa->narg <= sizeof(sa->args) / sizeof(sa->args[0]),
 877             ("Too many syscall arguments!"));
 878         error = 0;
 879         argp = &frame->tf_rdi;
 880         argp += reg;
 881         bcopy(argp, sa->args, sizeof(sa->args[0]) * regcnt);
 882         if (sa->narg > regcnt) {
 883                 KASSERT(params != NULL, ("copyin args with no params!"));
 884                 error = copyin(params, &sa->args[regcnt],
 885                     (sa->narg - regcnt) * sizeof(sa->args[0]));
 886         }
 887
 888         if (error == 0) {
 889                 td->td_retval[0] = 0;
 890                 td->td_retval[1] = frame->tf_rdx;
 891         }
 892
 893         return (error);
 894 }
 895
 896 #include "../../kern/subr_syscall.c"
 897
 898 /*
 899  * System call handler for native binaries.  The trap frame is already
 900  * set up by the assembler trampoline and a pointer to it is saved in
 901  * td_frame.
 902  */
 903 void
 904 amd64_syscall(struct thread *td, int traced)
 905 {
 906         int error;
 907         ksiginfo_t ksi;
 908
 909 #ifdef DIAGNOSTIC
 910         if (!TRAPF_USERMODE(td->td_frame)) {
 911                 panic("syscall");
 912                 /* NOT REACHED */
 913         }
 914 #endif
 915         error = syscallenter(td);
 916
 917         /*
 918          * Traced syscall.
 919          */
 920         if (__predict_false(traced)) {
 921                 td->td_frame->tf_rflags &= ~PSL_T;
 922                 ksiginfo_init_trap(&ksi);
 923                 ksi.ksi_signo = SIGTRAP;
 924                 ksi.ksi_code = TRAP_TRACE;
 925                 ksi.ksi_addr = (void *)td->td_frame->tf_rip;
 926                 trapsignal(td, &ksi);
 927         }
 928
 929         KASSERT(PCB_USER_FPU(td->td_pcb),
 930             ("System call %s returning with kernel FPU ctx leaked",
 931              syscallname(td->td_proc, td->td_sa.code)));
 932         KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td),
 933             ("System call %s returning with mangled pcb_save",
 934              syscallname(td->td_proc, td->td_sa.code)));
 935         KASSERT(td->td_md.md_invl_gen.gen == 0,
 936             ("System call %s returning with leaked invl_gen %lu",
 937             syscallname(td->td_proc, td->td_sa.code),
 938             td->td_md.md_invl_gen.gen));
 939
 940         syscallret(td, error);
 941
 942         /*
 943          * If the user-supplied value of %rip is not a canonical
 944          * address, then some CPUs will trigger a ring 0 #GP during
 945          * the sysret instruction.  However, the fault handler would
 946          * execute in ring 0 with the user's %gs and %rsp which would
 947          * not be safe.  Instead, use the full return path which
 948          * catches the problem safely.
 949          */
 950         if (__predict_false(td->td_frame->tf_rip >= VM_MAXUSER_ADDRESS))
 951                 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
 952 }