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;
 180 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
 181     &prot_fault_translation, 0,
 182     "Select signal to deliver on protection fault");
 183 static int uprintf_signal;
 184 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW,
 185     &uprintf_signal, 0,
 186     "Print debugging information on trap signal to ctty");
 187
 188 /*
 189  * Exception, fault, and trap interface to the FreeBSD kernel.
 190  * This common code is called from assembly language IDT gate entry
 191  * routines that prepare a suitable stack frame, and restore this
 192  * frame after the exception has been processed.
 193  */
 194
 195 void
 196 trap(struct trapframe *frame)
 197 {
 198         struct thread *td = curthread;
 199         struct proc *p = td->td_proc;
 200         int i = 0, ucode = 0, code;
 201         u_int type;
 202         register_t addr = 0;
 203         ksiginfo_t ksi;
 204
 205         PCPU_INC(cnt.v_trap);
 206         type = frame->tf_trapno;
 207
 208 #ifdef SMP
 209         /* Handler for NMI IPIs used for stopping CPUs. */
 210         if (type == T_NMI) {
 211                  if (ipi_nmi_handler() == 0)
 212                            goto out;
 213         }
 214 #endif /* SMP */
 215
 216 #ifdef KDB
 217         if (kdb_active) {
 218                 kdb_reenter();
 219                 goto out;
 220         }
 221 #endif
 222
 223         if (type == T_RESERVED) {
 224                 trap_fatal(frame, 0);
 225                 goto out;
 226         }
 227
 228 #ifdef  HWPMC_HOOKS
 229         /*
 230          * CPU PMCs interrupt using an NMI.  If the PMC module is
 231          * active, pass the 'rip' value to the PMC module's interrupt
 232          * handler.  A return value of '1' from the handler means that
 233          * the NMI was handled by it and we can return immediately.
 234          */
 235         if (type == T_NMI && pmc_intr &&
 236             (*pmc_intr)(PCPU_GET(cpuid), frame))
 237                 goto out;
 238 #endif
 239
 240         if (type == T_MCHK) {
 241                 mca_intr();
 242                 goto out;
 243         }
 244
 245 #ifdef KDTRACE_HOOKS
 246         /*
 247          * A trap can occur while DTrace executes a probe. Before
 248          * executing the probe, DTrace blocks re-scheduling and sets
 249          * a flag in it's per-cpu flags to indicate that it doesn't
 250          * want to fault. On returning from the probe, the no-fault
 251          * flag is cleared and finally re-scheduling is enabled.
 252          *
 253          * If the DTrace kernel module has registered a trap handler,
 254          * call it and if it returns non-zero, assume that it has
 255          * handled the trap and modified the trap frame so that this
 256          * function can return normally.
 257          */
 258         if (type == T_DTRACE_PROBE || type == T_DTRACE_RET ||
 259             type == T_BPTFLT) {
 260                 struct reg regs;
 261
 262                 fill_frame_regs(frame, &regs);
 263                 if (type == T_DTRACE_PROBE &&
 264                     dtrace_fasttrap_probe_ptr != NULL &&
 265                     dtrace_fasttrap_probe_ptr(&regs) == 0)
 266                         goto out;
 267                 else if (type == T_BPTFLT &&
 268                     dtrace_pid_probe_ptr != NULL &&
 269                     dtrace_pid_probe_ptr(&regs) == 0)
 270                         goto out;
 271                 else if (type == T_DTRACE_RET &&
 272                     dtrace_return_probe_ptr != NULL &&
 273                     dtrace_return_probe_ptr(&regs) == 0)
 274                         goto out;
 275         }
 276         if (dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
 277                 goto out;
 278 #endif
 279
 280         if ((frame->tf_rflags & PSL_I) == 0) {
 281                 /*
 282                  * Buggy application or kernel code has disabled
 283                  * interrupts and then trapped.  Enabling interrupts
 284                  * now is wrong, but it is better than running with
 285                  * interrupts disabled until they are accidentally
 286                  * enabled later.
 287                  */
 288                 if (ISPL(frame->tf_cs) == SEL_UPL)
 289                         uprintf(
 290                             "pid %ld (%s): trap %d with interrupts disabled\n",
 291                             (long)curproc->p_pid, curthread->td_name, type);
 292                 else if (type != T_NMI && type != T_BPTFLT &&
 293                     type != T_TRCTRAP) {
 294                         /*
 295                          * XXX not quite right, since this may be for a
 296                          * multiple fault in user mode.
 297                          */
 298                         printf("kernel trap %d with interrupts disabled\n",
 299                             type);
 300
 301                         /*
 302                          * We shouldn't enable interrupts while holding a
 303                          * spin lock.
 304                          */
 305                         if (td->td_md.md_spinlock_count == 0)
 306                                 enable_intr();
 307                 }
 308         }
 309
 310         code = frame->tf_err;
 311
 312         if (ISPL(frame->tf_cs) == SEL_UPL) {
 313                 /* user trap */
 314
 315                 td->td_pticks = 0;
 316                 td->td_frame = frame;
 317                 addr = frame->tf_rip;
 318                 if (td->td_ucred != p->p_ucred)
 319                         cred_update_thread(td);
 320
 321                 switch (type) {
 322                 case T_PRIVINFLT:       /* privileged instruction fault */
 323                         i = SIGILL;
 324                         ucode = ILL_PRVOPC;
 325                         break;
 326
 327                 case T_BPTFLT:          /* bpt instruction fault */
 328                 case T_TRCTRAP:         /* trace trap */
 329                         enable_intr();
 330                         frame->tf_rflags &= ~PSL_T;
 331                         i = SIGTRAP;
 332                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 333                         break;
 334
 335                 case T_ARITHTRAP:       /* arithmetic trap */
 336                         ucode = fputrap_x87();
 337                         if (ucode == -1)
 338                                 goto userout;
 339                         i = SIGFPE;
 340                         break;
 341
 342                 case T_PROTFLT:         /* general protection fault */
 343                         i = SIGBUS;
 344                         ucode = BUS_OBJERR;
 345                         break;
 346                 case T_STKFLT:          /* stack fault */
 347                 case T_SEGNPFLT:        /* segment not present fault */
 348                         i = SIGBUS;
 349                         ucode = BUS_ADRERR;
 350                         break;
 351                 case T_TSSFLT:          /* invalid TSS fault */
 352                         i = SIGBUS;
 353                         ucode = BUS_OBJERR;
 354                         break;
 355                 case T_DOUBLEFLT:       /* double fault */
 356                 default:
 357                         i = SIGBUS;
 358                         ucode = BUS_OBJERR;
 359                         break;
 360
 361                 case T_PAGEFLT:         /* page fault */
 362                         addr = frame->tf_addr;
 363                         i = trap_pfault(frame, TRUE);
 364                         if (i == -1)
 365                                 goto userout;
 366                         if (i == 0)
 367                                 goto user;
 368
 369                         if (i == SIGSEGV)
 370                                 ucode = SEGV_MAPERR;
 371                         else {
 372                                 if (prot_fault_translation == 0) {
 373                                         /*
 374                                          * Autodetect.
 375                                          * This check also covers the images
 376                                          * without the ABI-tag ELF note.
 377                                          */
 378                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 379                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 380                                                 i = SIGSEGV;
 381                                                 ucode = SEGV_ACCERR;
 382                                         } else {
 383                                                 i = SIGBUS;
 384                                                 ucode = BUS_PAGE_FAULT;
 385                                         }
 386                                 } else if (prot_fault_translation == 1) {
 387                                         /*
 388                                          * Always compat mode.
 389                                          */
 390                                         i = SIGBUS;
 391                                         ucode = BUS_PAGE_FAULT;
 392                                 } else {
 393                                         /*
 394                                          * Always SIGSEGV mode.
 395                                          */
 396                                         i = SIGSEGV;
 397                                         ucode = SEGV_ACCERR;
 398                                 }
 399                         }
 400                         break;
 401
 402                 case T_DIVIDE:          /* integer divide fault */
 403                         ucode = FPE_INTDIV;
 404                         i = SIGFPE;
 405                         break;
 406
 407 #ifdef DEV_ISA
 408                 case T_NMI:
 409                         /* machine/parity/power fail/"kitchen sink" faults */
 410                         if (isa_nmi(code) == 0) {
 411 #ifdef KDB
 412                                 /*
 413                                  * NMI can be hooked up to a pushbutton
 414                                  * for debugging.
 415                                  */
 416                                 if (kdb_on_nmi) {
 417                                         printf ("NMI ... going to debugger\n");
 418                                         kdb_trap(type, 0, frame);
 419                                 }
 420 #endif /* KDB */
 421                                 goto userout;
 422                         } else if (panic_on_nmi)
 423                                 panic("NMI indicates hardware failure");
 424                         break;
 425 #endif /* DEV_ISA */
 426
 427                 case T_OFLOW:           /* integer overflow fault */
 428                         ucode = FPE_INTOVF;
 429                         i = SIGFPE;
 430                         break;
 431
 432                 case T_BOUND:           /* bounds check fault */
 433                         ucode = FPE_FLTSUB;
 434                         i = SIGFPE;
 435                         break;
 436
 437                 case T_DNA:
 438                         /* transparent fault (due to context switch "late") */
 439                         KASSERT(PCB_USER_FPU(td->td_pcb),
 440                             ("kernel FPU ctx has leaked"));
 441                         fpudna();
 442                         goto userout;
 443
 444                 case T_FPOPFLT:         /* FPU operand fetch fault */
 445                         ucode = ILL_COPROC;
 446                         i = SIGILL;
 447                         break;
 448
 449                 case T_XMMFLT:          /* SIMD floating-point exception */
 450                         ucode = fputrap_sse();
 451                         if (ucode == -1)
 452                                 goto userout;
 453                         i = SIGFPE;
 454                         break;
 455                 }
 456         } else {
 457                 /* kernel trap */
 458
 459                 KASSERT(cold || td->td_ucred != NULL,
 460                     ("kernel trap doesn't have ucred"));
 461                 switch (type) {
 462                 case T_PAGEFLT:                 /* page fault */
 463                         (void) trap_pfault(frame, FALSE);
 464                         goto out;
 465
 466                 case T_DNA:
 467                         KASSERT(!PCB_USER_FPU(td->td_pcb),
 468                             ("Unregistered use of FPU in kernel"));
 469                         fpudna();
 470                         goto out;
 471
 472                 case T_ARITHTRAP:       /* arithmetic trap */
 473                 case T_XMMFLT:          /* SIMD floating-point exception */
 474                 case T_FPOPFLT:         /* FPU operand fetch fault */
 475                         /*
 476                          * XXXKIB for now disable any FPU traps in kernel
 477                          * handler registration seems to be overkill
 478                          */
 479                         trap_fatal(frame, 0);
 480                         goto out;
 481
 482                 case T_STKFLT:          /* stack fault */
 483                         break;
 484
 485                 case T_PROTFLT:         /* general protection fault */
 486                 case T_SEGNPFLT:        /* segment not present fault */
 487                         if (td->td_intr_nesting_level != 0)
 488                                 break;
 489
 490                         /*
 491                          * Invalid segment selectors and out of bounds
 492                          * %rip's and %rsp's can be set up in user mode.
 493                          * This causes a fault in kernel mode when the
 494                          * kernel tries to return to user mode.  We want
 495                          * to get this fault so that we can fix the
 496                          * problem here and not have to check all the
 497                          * selectors and pointers when the user changes
 498                          * them.
 499                          */
 500                         if (frame->tf_rip == (long)doreti_iret) {
 501                                 frame->tf_rip = (long)doreti_iret_fault;
 502                                 goto out;
 503                         }
 504                         if (frame->tf_rip == (long)ld_ds) {
 505                                 frame->tf_rip = (long)ds_load_fault;
 506                                 goto out;
 507                         }
 508                         if (frame->tf_rip == (long)ld_es) {
 509                                 frame->tf_rip = (long)es_load_fault;
 510                                 goto out;
 511                         }
 512                         if (frame->tf_rip == (long)ld_fs) {
 513                                 frame->tf_rip = (long)fs_load_fault;
 514                                 goto out;
 515                         }
 516                         if (frame->tf_rip == (long)ld_gs) {
 517                                 frame->tf_rip = (long)gs_load_fault;
 518                                 goto out;
 519                         }
 520                         if (frame->tf_rip == (long)ld_gsbase) {
 521                                 frame->tf_rip = (long)gsbase_load_fault;
 522                                 goto out;
 523                         }
 524                         if (frame->tf_rip == (long)ld_fsbase) {
 525                                 frame->tf_rip = (long)fsbase_load_fault;
 526                                 goto out;
 527                         }
 528                         if (curpcb->pcb_onfault != NULL) {
 529                                 frame->tf_rip = (long)curpcb->pcb_onfault;
 530                                 goto out;
 531                         }
 532                         break;
 533
 534                 case T_TSSFLT:
 535                         /*
 536                          * PSL_NT can be set in user mode and isn't cleared
 537                          * automatically when the kernel is entered.  This
 538                          * causes a TSS fault when the kernel attempts to
 539                          * `iret' because the TSS link is uninitialized.  We
 540                          * want to get this fault so that we can fix the
 541                          * problem here and not every time the kernel is
 542                          * entered.
 543                          */
 544                         if (frame->tf_rflags & PSL_NT) {
 545                                 frame->tf_rflags &= ~PSL_NT;
 546                                 goto out;
 547                         }
 548                         break;
 549
 550                 case T_TRCTRAP:  /* trace trap */
 551                         /*
 552                          * Ignore debug register trace traps due to
 553                          * accesses in the user's address space, which
 554                          * can happen under several conditions such as
 555                          * if a user sets a watchpoint on a buffer and
 556                          * then passes that buffer to a system call.
 557                          * We still want to get TRCTRAPS for addresses
 558                          * in kernel space because that is useful when
 559                          * debugging the kernel.
 560                          */
 561                         if (user_dbreg_trap()) {
 562                                 /*
 563                                  * Reset breakpoint bits because the
 564                                  * processor doesn't
 565                                  */
 566                                 /* XXX check upper bits here */
 567                                 load_dr6(rdr6() & 0xfffffff0);
 568                                 goto out;
 569                         }
 570                         /*
 571                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 572                          */
 573                 case T_BPTFLT:
 574                         /*
 575                          * If KDB is enabled, let it handle the debugger trap.
 576                          * Otherwise, debugger traps "can't happen".
 577                          */
 578 #ifdef KDB
 579                         if (kdb_trap(type, 0, frame))
 580                                 goto out;
 581 #endif
 582                         break;
 583
 584 #ifdef DEV_ISA
 585                 case T_NMI:
 586                         /* machine/parity/power fail/"kitchen sink" faults */
 587                         if (isa_nmi(code) == 0) {
 588 #ifdef KDB
 589                                 /*
 590                                  * NMI can be hooked up to a pushbutton
 591                                  * for debugging.
 592                                  */
 593                                 if (kdb_on_nmi) {
 594                                         printf ("NMI ... going to debugger\n");
 595                                         kdb_trap(type, 0, frame);
 596                                 }
 597 #endif /* KDB */
 598                                 goto out;
 599                         } else if (panic_on_nmi == 0)
 600                                 goto out;
 601                         /* FALLTHROUGH */
 602 #endif /* DEV_ISA */
 603                 }
 604
 605                 trap_fatal(frame, 0);
 606                 goto out;
 607         }
 608
 609         /* Translate fault for emulators (e.g. Linux) */
 610         if (*p->p_sysent->sv_transtrap)
 611                 i = (*p->p_sysent->sv_transtrap)(i, type);
 612
 613         ksiginfo_init_trap(&ksi);
 614         ksi.ksi_signo = i;
 615         ksi.ksi_code = ucode;
 616         ksi.ksi_trapno = type;
 617         ksi.ksi_addr = (void *)addr;
 618         if (uprintf_signal) {
 619                 uprintf("pid %d comm %s: signal %d err %lx code %d type %d "
 620                     "addr 0x%lx rsp 0x%lx rip 0x%lx "
 621                     "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
 622                     p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr,
 623                     frame->tf_rsp, frame->tf_rip,
 624                     fubyte((void *)(frame->tf_rip + 0)),
 625                     fubyte((void *)(frame->tf_rip + 1)),
 626                     fubyte((void *)(frame->tf_rip + 2)),
 627                     fubyte((void *)(frame->tf_rip + 3)),
 628                     fubyte((void *)(frame->tf_rip + 4)),
 629                     fubyte((void *)(frame->tf_rip + 5)),
 630                     fubyte((void *)(frame->tf_rip + 6)),
 631                     fubyte((void *)(frame->tf_rip + 7)));
 632         }
 633         KASSERT((read_rflags() & PSL_I) != 0, ("interrupts disabled"));
 634         trapsignal(td, &ksi);
 635
 636 user:
 637         userret(td, frame);
 638         mtx_assert(&Giant, MA_NOTOWNED);
 639         KASSERT(PCB_USER_FPU(td->td_pcb),
 640             ("Return from trap with kernel FPU ctx leaked"));
 641 userout:
 642 out:
 643         return;
 644 }
 645
 646 static int
 647 trap_pfault(frame, usermode)
 648         struct trapframe *frame;
 649         int usermode;
 650 {
 651         vm_offset_t va;
 652         struct vmspace *vm = NULL;
 653         vm_map_t map;
 654         int rv = 0;
 655         vm_prot_t ftype;
 656         struct thread *td = curthread;
 657         struct proc *p = td->td_proc;
 658         vm_offset_t eva = frame->tf_addr;
 659
 660         if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
 661                 /*
 662                  * Due to both processor errata and lazy TLB invalidation when
 663                  * access restrictions are removed from virtual pages, memory
 664                  * accesses that are allowed by the physical mapping layer may
 665                  * nonetheless cause one spurious page fault per virtual page.
 666                  * When the thread is executing a "no faulting" section that
 667                  * is bracketed by vm_fault_{disable,enable}_pagefaults(),
 668                  * every page fault is treated as a spurious page fault,
 669                  * unless it accesses the same virtual address as the most
 670                  * recent page fault within the same "no faulting" section.
 671                  */
 672                 if (td->td_md.md_spurflt_addr != eva ||
 673                     (td->td_pflags & TDP_RESETSPUR) != 0) {
 674                         /*
 675                          * Do nothing to the TLB.  A stale TLB entry is
 676                          * flushed automatically by a page fault.
 677                          */
 678                         td->td_md.md_spurflt_addr = eva;
 679                         td->td_pflags &= ~TDP_RESETSPUR;
 680                         return (0);
 681                 }
 682         } else {
 683                 /*
 684                  * If we get a page fault while in a critical section, then
 685                  * it is most likely a fatal kernel page fault.  The kernel
 686                  * is already going to panic trying to get a sleep lock to
 687                  * do the VM lookup, so just consider it a fatal trap so the
 688                  * kernel can print out a useful trap message and even get
 689                  * to the debugger.
 690                  *
 691                  * If we get a page fault while holding a non-sleepable
 692                  * lock, then it is most likely a fatal kernel page fault.
 693                  * If WITNESS is enabled, then it's going to whine about
 694                  * bogus LORs with various VM locks, so just skip to the
 695                  * fatal trap handling directly.
 696                  */
 697                 if (td->td_critnest != 0 ||
 698                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 699                     "Kernel page fault") != 0) {
 700                         trap_fatal(frame, eva);
 701                         return (-1);
 702                 }
 703         }
 704         va = trunc_page(eva);
 705         if (va >= VM_MIN_KERNEL_ADDRESS) {
 706                 /*
 707                  * Don't allow user-mode faults in kernel address space.
 708                  */
 709                 if (usermode)
 710                         goto nogo;
 711
 712                 map = kernel_map;
 713         } else {
 714                 /*
 715                  * This is a fault on non-kernel virtual memory.
 716                  * vm is initialized above to NULL. If curproc is NULL
 717                  * or curproc->p_vmspace is NULL the fault is fatal.
 718                  */
 719                 if (p != NULL)
 720                         vm = p->p_vmspace;
 721
 722                 if (vm == NULL)
 723                         goto nogo;
 724
 725                 map = &vm->vm_map;
 726
 727                 /*
 728                  * When accessing a usermode address, kernel must be
 729                  * ready to accept the page fault, and provide a
 730                  * handling routine.  Since accessing the address
 731                  * without the handler is a bug, do not try to handle
 732                  * it normally, and panic immediately.
 733                  */
 734                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 735                     curpcb->pcb_onfault == NULL)) {
 736                         trap_fatal(frame, eva);
 737                         return (-1);
 738                 }
 739         }
 740
 741         /*
 742          * PGEX_I is defined only if the execute disable bit capability is
 743          * supported and enabled.
 744          */
 745         if (frame->tf_err & PGEX_W)
 746                 ftype = VM_PROT_WRITE;
 747         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 748                 ftype = VM_PROT_EXECUTE;
 749         else
 750                 ftype = VM_PROT_READ;
 751
 752         if (map != kernel_map) {
 753                 /*
 754                  * Keep swapout from messing with us during this
 755                  *      critical time.
 756                  */
 757                 PROC_LOCK(p);
 758                 ++p->p_lock;
 759                 PROC_UNLOCK(p);
 760
 761                 /* Fault in the user page: */
 762                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 763
 764                 PROC_LOCK(p);
 765                 --p->p_lock;
 766                 PROC_UNLOCK(p);
 767         } else {
 768                 /*
 769                  * Don't have to worry about process locking or stacks in the
 770                  * kernel.
 771                  */
 772                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 773         }
 774         if (rv == KERN_SUCCESS) {
 775 #ifdef HWPMC_HOOKS
 776                 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
 777                         PMC_SOFT_CALL_TF( , , page_fault, all, frame);
 778                         if (ftype == VM_PROT_READ)
 779                                 PMC_SOFT_CALL_TF( , , page_fault, read,
 780                                     frame);
 781                         else
 782                                 PMC_SOFT_CALL_TF( , , page_fault, write,
 783                                     frame);
 784                 }
 785 #endif
 786                 return (0);
 787         }
 788 nogo:
 789         if (!usermode) {
 790                 if (td->td_intr_nesting_level == 0 &&
 791                     curpcb->pcb_onfault != NULL) {
 792                         frame->tf_rip = (long)curpcb->pcb_onfault;
 793                         return (0);
 794                 }
 795                 trap_fatal(frame, eva);
 796                 return (-1);
 797         }
 798
 799         return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 800 }
 801
 802 static void
 803 trap_fatal(frame, eva)
 804         struct trapframe *frame;
 805         vm_offset_t eva;
 806 {
 807         int code, ss;
 808         u_int type;
 809         long esp;
 810         struct soft_segment_descriptor softseg;
 811         char *msg;
 812
 813         code = frame->tf_err;
 814         type = frame->tf_trapno;
 815         sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)],
 816             &softseg);
 817
 818         if (type <= MAX_TRAP_MSG)
 819                 msg = trap_msg[type];
 820         else
 821                 msg = "UNKNOWN";
 822         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 823             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
 824 #ifdef SMP
 825         /* two separate prints in case of a trap on an unmapped page */
 826         printf("cpuid = %d; ", PCPU_GET(cpuid));
 827         printf("apic id = %02x\n", PCPU_GET(apic_id));
 828 #endif
 829         if (type == T_PAGEFLT) {
 830                 printf("fault virtual address   = 0x%lx\n", eva);
 831                 printf("fault code              = %s %s %s, %s\n",
 832                         code & PGEX_U ? "user" : "supervisor",
 833                         code & PGEX_W ? "write" : "read",
 834                         code & PGEX_I ? "instruction" : "data",
 835                         code & PGEX_P ? "protection violation" : "page not present");
 836         }
 837         printf("instruction pointer     = 0x%lx:0x%lx\n",
 838                frame->tf_cs & 0xffff, frame->tf_rip);
 839         if (ISPL(frame->tf_cs) == SEL_UPL) {
 840                 ss = frame->tf_ss & 0xffff;
 841                 esp = frame->tf_rsp;
 842         } else {
 843                 ss = GSEL(GDATA_SEL, SEL_KPL);
 844                 esp = (long)&frame->tf_rsp;
 845         }
 846         printf("stack pointer           = 0x%x:0x%lx\n", ss, esp);
 847         printf("frame pointer           = 0x%x:0x%lx\n", ss, frame->tf_rbp);
 848         printf("code segment            = base 0x%lx, limit 0x%lx, type 0x%x\n",
 849                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 850         printf("                        = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
 851                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
 852                softseg.ssd_gran);
 853         printf("processor eflags        = ");
 854         if (frame->tf_rflags & PSL_T)
 855                 printf("trace trap, ");
 856         if (frame->tf_rflags & PSL_I)
 857                 printf("interrupt enabled, ");
 858         if (frame->tf_rflags & PSL_NT)
 859                 printf("nested task, ");
 860         if (frame->tf_rflags & PSL_RF)
 861                 printf("resume, ");
 862         printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
 863         printf("current process         = ");
 864         if (curproc) {
 865                 printf("%lu (%s)\n",
 866                     (u_long)curproc->p_pid, curthread->td_name ?
 867                     curthread->td_name : "");
 868         } else {
 869                 printf("Idle\n");
 870         }
 871
 872 #ifdef KDB
 873         if (debugger_on_panic || kdb_active)
 874                 if (kdb_trap(type, 0, frame))
 875                         return;
 876 #endif
 877         printf("trap number             = %d\n", type);
 878         if (type <= MAX_TRAP_MSG)
 879                 panic("%s", trap_msg[type]);
 880         else
 881                 panic("unknown/reserved trap");
 882 }
 883
 884 /*
 885  * Double fault handler. Called when a fault occurs while writing
 886  * a frame for a trap/exception onto the stack. This usually occurs
 887  * when the stack overflows (such is the case with infinite recursion,
 888  * for example).
 889  */
 890 void
 891 dblfault_handler(struct trapframe *frame)
 892 {
 893 #ifdef KDTRACE_HOOKS
 894         if (dtrace_doubletrap_func != NULL)
 895                 (*dtrace_doubletrap_func)();
 896 #endif
 897         printf("\nFatal double fault\n");
 898         printf("rip = 0x%lx\n", frame->tf_rip);
 899         printf("rsp = 0x%lx\n", frame->tf_rsp);
 900         printf("rbp = 0x%lx\n", frame->tf_rbp);
 901 #ifdef SMP
 902         /* two separate prints in case of a trap on an unmapped page */
 903         printf("cpuid = %d; ", PCPU_GET(cpuid));
 904         printf("apic id = %02x\n", PCPU_GET(apic_id));
 905 #endif
 906         panic("double fault");
 907 }
 908
 909 int
 910 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
 911 {
 912         struct proc *p;
 913         struct trapframe *frame;
 914         register_t *argp;
 915         caddr_t params;
 916         int reg, regcnt, error;
 917
 918         p = td->td_proc;
 919         frame = td->td_frame;
 920         reg = 0;
 921         regcnt = 6;
 922
 923         params = (caddr_t)frame->tf_rsp + sizeof(register_t);
 924         sa->code = frame->tf_rax;
 925
 926         if (sa->code == SYS_syscall || sa->code == SYS___syscall) {
 927                 sa->code = frame->tf_rdi;
 928                 reg++;
 929                 regcnt--;
 930         }
 931         if (p->p_sysent->sv_mask)
 932                 sa->code &= p->p_sysent->sv_mask;
 933
 934         if (sa->code >= p->p_sysent->sv_size)
 935                 sa->callp = &p->p_sysent->sv_table[0];
 936         else
 937                 sa->callp = &p->p_sysent->sv_table[sa->code];
 938
 939         sa->narg = sa->callp->sy_narg;
 940         KASSERT(sa->narg <= sizeof(sa->args) / sizeof(sa->args[0]),
 941             ("Too many syscall arguments!"));
 942         error = 0;
 943         argp = &frame->tf_rdi;
 944         argp += reg;
 945         bcopy(argp, sa->args, sizeof(sa->args[0]) * regcnt);
 946         if (sa->narg > regcnt) {
 947                 KASSERT(params != NULL, ("copyin args with no params!"));
 948                 error = copyin(params, &sa->args[regcnt],
 949                     (sa->narg - regcnt) * sizeof(sa->args[0]));
 950         }
 951
 952         if (error == 0) {
 953                 td->td_retval[0] = 0;
 954                 td->td_retval[1] = frame->tf_rdx;
 955         }
 956
 957         return (error);
 958 }
 959
 960 #include "../../kern/subr_syscall.c"
 961
 962 /*
 963  * System call handler for native binaries.  The trap frame is already
 964  * set up by the assembler trampoline and a pointer to it is saved in
 965  * td_frame.
 966  */
 967 void
 968 amd64_syscall(struct thread *td, int traced)
 969 {
 970         struct syscall_args sa;
 971         int error;
 972         ksiginfo_t ksi;
 973
 974 #ifdef DIAGNOSTIC
 975         if (ISPL(td->td_frame->tf_cs) != SEL_UPL) {
 976                 panic("syscall");
 977                 /* NOT REACHED */
 978         }
 979 #endif
 980         error = syscallenter(td, &sa);
 981
 982         /*
 983          * Traced syscall.
 984          */
 985         if (__predict_false(traced)) {
 986                 td->td_frame->tf_rflags &= ~PSL_T;
 987                 ksiginfo_init_trap(&ksi);
 988                 ksi.ksi_signo = SIGTRAP;
 989                 ksi.ksi_code = TRAP_TRACE;
 990                 ksi.ksi_addr = (void *)td->td_frame->tf_rip;
 991                 trapsignal(td, &ksi);
 992         }
 993
 994         KASSERT(PCB_USER_FPU(td->td_pcb),
 995             ("System call %s returing with kernel FPU ctx leaked",
 996              syscallname(td->td_proc, sa.code)));
 997         KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td),
 998             ("System call %s returning with mangled pcb_save",
 999              syscallname(td->td_proc, sa.code)));
1000
1001         syscallret(td, error, &sa);
1002
1003         /*
1004          * If the user-supplied value of %rip is not a canonical
1005          * address, then some CPUs will trigger a ring 0 #GP during
1006          * the sysret instruction.  However, the fault handler would
1007          * execute in ring 0 with the user's %gs and %rsp which would
1008          * not be safe.  Instead, use the full return path which
1009          * catches the problem safely.
1010          */
1011         if (td->td_frame->tf_rip >= VM_MAXUSER_ADDRESS)
1012                 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
1013 }