sys/i386/i386/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  * 386 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 #include "opt_ktrace.h"
  54 #include "opt_npx.h"
  55 #include "opt_trap.h"
  56
  57 #include <sys/param.h>
  58 #include <sys/bus.h>
  59 #include <sys/systm.h>
  60 #include <sys/proc.h>
  61 #include <sys/pioctl.h>
  62 #include <sys/ptrace.h>
  63 #include <sys/kdb.h>
  64 #include <sys/kernel.h>
  65 #include <sys/ktr.h>
  66 #include <sys/lock.h>
  67 #include <sys/mutex.h>
  68 #include <sys/resourcevar.h>
  69 #include <sys/signalvar.h>
  70 #include <sys/syscall.h>
  71 #include <sys/sysctl.h>
  72 #include <sys/sysent.h>
  73 #include <sys/uio.h>
  74 #include <sys/vmmeter.h>
  75 #ifdef KTRACE
  76 #include <sys/ktrace.h>
  77 #endif
  78 #ifdef HWPMC_HOOKS
  79 #include <sys/pmckern.h>
  80 #endif
  81 #include <security/audit/audit.h>
  82
  83 #include <vm/vm.h>
  84 #include <vm/vm_param.h>
  85 #include <vm/pmap.h>
  86 #include <vm/vm_kern.h>
  87 #include <vm/vm_map.h>
  88 #include <vm/vm_page.h>
  89 #include <vm/vm_extern.h>
  90
  91 #include <machine/cpu.h>
  92 #include <machine/intr_machdep.h>
  93 #include <machine/mca.h>
  94 #include <machine/md_var.h>
  95 #include <machine/pcb.h>
  96 #ifdef SMP
  97 #include <machine/smp.h>
  98 #endif
  99 #include <machine/tss.h>
 100 #include <machine/vm86.h>
 101
 102 #ifdef POWERFAIL_NMI
 103 #include <sys/syslog.h>
 104 #include <machine/clock.h>
 105 #endif
 106
 107 #ifdef KDTRACE_HOOKS
 108 #include <sys/dtrace_bsd.h>
 109
 110 /*
 111  * This is a hook which is initialised by the dtrace module
 112  * to handle traps which might occur during DTrace probe
 113  * execution.
 114  */
 115 dtrace_trap_func_t      dtrace_trap_func;
 116
 117 dtrace_doubletrap_func_t        dtrace_doubletrap_func;
 118
 119 /*
 120  * This is a hook which is initialised by the systrace module
 121  * when it is loaded. This keeps the DTrace syscall provider
 122  * implementation opaque.
 123  */
 124 systrace_probe_func_t   systrace_probe_func;
 125
 126 /*
 127  * These hooks are necessary for the pid, usdt and fasttrap providers.
 128  */
 129 dtrace_fasttrap_probe_ptr_t     dtrace_fasttrap_probe_ptr;
 130 dtrace_pid_probe_ptr_t          dtrace_pid_probe_ptr;
 131 dtrace_return_probe_ptr_t       dtrace_return_probe_ptr;
 132 #endif
 133
 134 extern void trap(struct trapframe *frame);
 135 extern void syscall(struct trapframe *frame);
 136
 137 static int trap_pfault(struct trapframe *, int, vm_offset_t);
 138 static void trap_fatal(struct trapframe *, vm_offset_t);
 139 void dblfault_handler(void);
 140
 141 extern inthand_t IDTVEC(lcall_syscall);
 142
 143 #define MAX_TRAP_MSG            30
 144 static char *trap_msg[] = {
 145         "",                                     /*  0 unused */
 146         "privileged instruction fault",         /*  1 T_PRIVINFLT */
 147         "",                                     /*  2 unused */
 148         "breakpoint instruction fault",         /*  3 T_BPTFLT */
 149         "",                                     /*  4 unused */
 150         "",                                     /*  5 unused */
 151         "arithmetic trap",                      /*  6 T_ARITHTRAP */
 152         "",                                     /*  7 unused */
 153         "",                                     /*  8 unused */
 154         "general protection fault",             /*  9 T_PROTFLT */
 155         "trace trap",                           /* 10 T_TRCTRAP */
 156         "",                                     /* 11 unused */
 157         "page fault",                           /* 12 T_PAGEFLT */
 158         "",                                     /* 13 unused */
 159         "alignment fault",                      /* 14 T_ALIGNFLT */
 160         "",                                     /* 15 unused */
 161         "",                                     /* 16 unused */
 162         "",                                     /* 17 unused */
 163         "integer divide fault",                 /* 18 T_DIVIDE */
 164         "non-maskable interrupt trap",          /* 19 T_NMI */
 165         "overflow trap",                        /* 20 T_OFLOW */
 166         "FPU bounds check fault",               /* 21 T_BOUND */
 167         "FPU device not available",             /* 22 T_DNA */
 168         "double fault",                         /* 23 T_DOUBLEFLT */
 169         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
 170         "invalid TSS fault",                    /* 25 T_TSSFLT */
 171         "segment not present fault",            /* 26 T_SEGNPFLT */
 172         "stack fault",                          /* 27 T_STKFLT */
 173         "machine check trap",                   /* 28 T_MCHK */
 174         "SIMD floating-point exception",        /* 29 T_XMMFLT */
 175         "reserved (unknown) fault",             /* 30 T_RESERVED */
 176 };
 177
 178 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 179 extern int has_f00f_bug;
 180 #endif
 181
 182 #ifdef KDB
 183 static int kdb_on_nmi = 1;
 184 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
 185         &kdb_on_nmi, 0, "Go to KDB on NMI");
 186 #endif
 187 static int panic_on_nmi = 1;
 188 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
 189         &panic_on_nmi, 0, "Panic on NMI");
 190 static int prot_fault_translation = 0;
 191 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
 192         &prot_fault_translation, 0, "Select signal to deliver on protection fault");
 193
 194 /*
 195  * Exception, fault, and trap interface to the FreeBSD kernel.
 196  * This common code is called from assembly language IDT gate entry
 197  * routines that prepare a suitable stack frame, and restore this
 198  * frame after the exception has been processed.
 199  */
 200
 201 void
 202 trap(struct trapframe *frame)
 203 {
 204         struct thread *td = curthread;
 205         struct proc *p = td->td_proc;
 206         int i = 0, ucode = 0, code;
 207         u_int type;
 208         register_t addr = 0;
 209         vm_offset_t eva;
 210         ksiginfo_t ksi;
 211 #ifdef POWERFAIL_NMI
 212         static int lastalert = 0;
 213 #endif
 214
 215         PCPU_INC(cnt.v_trap);
 216         type = frame->tf_trapno;
 217
 218 #ifdef SMP
 219         /* Handler for NMI IPIs used for stopping CPUs. */
 220         if (type == T_NMI) {
 221                  if (ipi_nmi_handler() == 0)
 222                            goto out;
 223         }
 224 #endif /* SMP */
 225
 226 #ifdef KDB
 227         if (kdb_active) {
 228                 kdb_reenter();
 229                 goto out;
 230         }
 231 #endif
 232
 233         if (type == T_RESERVED) {
 234                 trap_fatal(frame, 0);
 235                 goto out;
 236         }
 237
 238 #ifdef  HWPMC_HOOKS
 239         /*
 240          * CPU PMCs interrupt using an NMI so we check for that first.
 241          * If the HWPMC module is active, 'pmc_hook' will point to
 242          * the function to be called.  A return value of '1' from the
 243          * hook means that the NMI was handled by it and that we can
 244          * return immediately.
 245          */
 246         if (type == T_NMI && pmc_intr &&
 247             (*pmc_intr)(PCPU_GET(cpuid), frame))
 248             goto out;
 249 #endif
 250
 251         if (type == T_MCHK) {
 252                 if (!mca_intr())
 253                         trap_fatal(frame, 0);
 254                 goto out;
 255         }
 256
 257 #ifdef KDTRACE_HOOKS
 258         /*
 259          * A trap can occur while DTrace executes a probe. Before
 260          * executing the probe, DTrace blocks re-scheduling and sets
 261          * a flag in it's per-cpu flags to indicate that it doesn't
 262          * want to fault. On returning from the probe, the no-fault
 263          * flag is cleared and finally re-scheduling is enabled.
 264          *
 265          * If the DTrace kernel module has registered a trap handler,
 266          * call it and if it returns non-zero, assume that it has
 267          * handled the trap and modified the trap frame so that this
 268          * function can return normally.
 269          */
 270         if ((type == T_PROTFLT || type == T_PAGEFLT) &&
 271             dtrace_trap_func != NULL)
 272                 if ((*dtrace_trap_func)(frame, type))
 273                         goto out;
 274         if (type == T_DTRACE_PROBE || type == T_DTRACE_RET ||
 275             type == T_BPTFLT) {
 276                 struct reg regs;
 277
 278                 regs.r_fs = frame->tf_fs;
 279                 regs.r_es = frame->tf_es;
 280                 regs.r_ds = frame->tf_ds;
 281                 regs.r_edi = frame->tf_edi;
 282                 regs.r_esi = frame->tf_esi;
 283                 regs.r_ebp = frame->tf_ebp;
 284                 regs.r_ebx = frame->tf_ebx;
 285                 regs.r_edx = frame->tf_edx;
 286                 regs.r_ecx = frame->tf_ecx;
 287                 regs.r_eax = frame->tf_eax;
 288                 regs.r_eip = frame->tf_eip;
 289                 regs.r_cs = frame->tf_cs;
 290                 regs.r_eflags = frame->tf_eflags;
 291                 regs.r_esp = frame->tf_esp;
 292                 regs.r_ss = frame->tf_ss;
 293                 if (type == T_DTRACE_PROBE &&
 294                     dtrace_fasttrap_probe_ptr != NULL &&
 295                     dtrace_fasttrap_probe_ptr(&regs) == 0)
 296                         goto out;
 297                 if (type == T_BPTFLT &&
 298                     dtrace_pid_probe_ptr != NULL &&
 299                     dtrace_pid_probe_ptr(&regs) == 0)
 300                         goto out;
 301                 if (type == T_DTRACE_RET &&
 302                     dtrace_return_probe_ptr != NULL &&
 303                     dtrace_return_probe_ptr(&regs) == 0)
 304                         goto out;
 305         }
 306 #endif
 307
 308         if ((frame->tf_eflags & PSL_I) == 0) {
 309                 /*
 310                  * Buggy application or kernel code has disabled
 311                  * interrupts and then trapped.  Enabling interrupts
 312                  * now is wrong, but it is better than running with
 313                  * interrupts disabled until they are accidentally
 314                  * enabled later.
 315                  */
 316                 if (ISPL(frame->tf_cs) == SEL_UPL || (frame->tf_eflags & PSL_VM))
 317                         uprintf(
 318                             "pid %ld (%s): trap %d with interrupts disabled\n",
 319                             (long)curproc->p_pid, curthread->td_name, type);
 320                 else if (type != T_BPTFLT && type != T_TRCTRAP &&
 321                          frame->tf_eip != (int)cpu_switch_load_gs) {
 322                         /*
 323                          * XXX not quite right, since this may be for a
 324                          * multiple fault in user mode.
 325                          */
 326                         printf("kernel trap %d with interrupts disabled\n",
 327                             type);
 328                         /*
 329                          * Page faults need interrupts disabled until later,
 330                          * and we shouldn't enable interrupts while holding
 331                          * a spin lock or if servicing an NMI.
 332                          */
 333                         if (type != T_NMI && type != T_PAGEFLT &&
 334                             td->td_md.md_spinlock_count == 0)
 335                                 enable_intr();
 336                 }
 337         }
 338         eva = 0;
 339         code = frame->tf_err;
 340         if (type == T_PAGEFLT) {
 341                 /*
 342                  * For some Cyrix CPUs, %cr2 is clobbered by
 343                  * interrupts.  This problem is worked around by using
 344                  * an interrupt gate for the pagefault handler.  We
 345                  * are finally ready to read %cr2 and then must
 346                  * reenable interrupts.
 347                  *
 348                  * If we get a page fault while in a critical section, then
 349                  * it is most likely a fatal kernel page fault.  The kernel
 350                  * is already going to panic trying to get a sleep lock to
 351                  * do the VM lookup, so just consider it a fatal trap so the
 352                  * kernel can print out a useful trap message and even get
 353                  * to the debugger.
 354                  *
 355                  * If we get a page fault while holding a non-sleepable
 356                  * lock, then it is most likely a fatal kernel page fault.
 357                  * If WITNESS is enabled, then it's going to whine about
 358                  * bogus LORs with various VM locks, so just skip to the
 359                  * fatal trap handling directly.
 360                  */
 361                 eva = rcr2();
 362                 if (td->td_critnest != 0 ||
 363                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 364                     "Kernel page fault") != 0)
 365                         trap_fatal(frame, eva);
 366                 else
 367                         enable_intr();
 368         }
 369
 370         if ((ISPL(frame->tf_cs) == SEL_UPL) ||
 371             ((frame->tf_eflags & PSL_VM) &&
 372                 !(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL))) {
 373                 /* user trap */
 374
 375                 td->td_pticks = 0;
 376                 td->td_frame = frame;
 377                 addr = frame->tf_eip;
 378                 if (td->td_ucred != p->p_ucred)
 379                         cred_update_thread(td);
 380
 381                 switch (type) {
 382                 case T_PRIVINFLT:       /* privileged instruction fault */
 383                         i = SIGILL;
 384                         ucode = ILL_PRVOPC;
 385                         break;
 386
 387                 case T_BPTFLT:          /* bpt instruction fault */
 388                 case T_TRCTRAP:         /* trace trap */
 389                         enable_intr();
 390                         frame->tf_eflags &= ~PSL_T;
 391                         i = SIGTRAP;
 392                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 393                         break;
 394
 395                 case T_ARITHTRAP:       /* arithmetic trap */
 396 #ifdef DEV_NPX
 397                         ucode = npxtrap();
 398                         if (ucode == -1)
 399                                 goto userout;
 400 #else
 401                         ucode = 0;
 402 #endif
 403                         i = SIGFPE;
 404                         break;
 405
 406                         /*
 407                          * The following two traps can happen in
 408                          * vm86 mode, and, if so, we want to handle
 409                          * them specially.
 410                          */
 411                 case T_PROTFLT:         /* general protection fault */
 412                 case T_STKFLT:          /* stack fault */
 413                         if (frame->tf_eflags & PSL_VM) {
 414                                 i = vm86_emulate((struct vm86frame *)frame);
 415                                 if (i == 0)
 416                                         goto user;
 417                                 break;
 418                         }
 419                         i = SIGBUS;
 420                         ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
 421                         break;
 422                 case T_SEGNPFLT:        /* segment not present fault */
 423                         i = SIGBUS;
 424                         ucode = BUS_ADRERR;
 425                         break;
 426                 case T_TSSFLT:          /* invalid TSS fault */
 427                         i = SIGBUS;
 428                         ucode = BUS_OBJERR;
 429                         break;
 430                 case T_DOUBLEFLT:       /* double fault */
 431                 default:
 432                         i = SIGBUS;
 433                         ucode = BUS_OBJERR;
 434                         break;
 435
 436                 case T_PAGEFLT:         /* page fault */
 437
 438                         i = trap_pfault(frame, TRUE, eva);
 439 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 440                         if (i == -2) {
 441                                 /*
 442                                  * The f00f hack workaround has triggered, so
 443                                  * treat the fault as an illegal instruction
 444                                  * (T_PRIVINFLT) instead of a page fault.
 445                                  */
 446                                 type = frame->tf_trapno = T_PRIVINFLT;
 447
 448                                 /* Proceed as in that case. */
 449                                 ucode = ILL_PRVOPC;
 450                                 i = SIGILL;
 451                                 break;
 452                         }
 453 #endif
 454                         if (i == -1)
 455                                 goto userout;
 456                         if (i == 0)
 457                                 goto user;
 458
 459                         if (i == SIGSEGV)
 460                                 ucode = SEGV_MAPERR;
 461                         else {
 462                                 if (prot_fault_translation == 0) {
 463                                         /*
 464                                          * Autodetect.
 465                                          * This check also covers the images
 466                                          * without the ABI-tag ELF note.
 467                                          */
 468                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 469                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 470                                                 i = SIGSEGV;
 471                                                 ucode = SEGV_ACCERR;
 472                                         } else {
 473                                                 i = SIGBUS;
 474                                                 ucode = BUS_PAGE_FAULT;
 475                                         }
 476                                 } else if (prot_fault_translation == 1) {
 477                                         /*
 478                                          * Always compat mode.
 479                                          */
 480                                         i = SIGBUS;
 481                                         ucode = BUS_PAGE_FAULT;
 482                                 } else {
 483                                         /*
 484                                          * Always SIGSEGV mode.
 485                                          */
 486                                         i = SIGSEGV;
 487                                         ucode = SEGV_ACCERR;
 488                                 }
 489                         }
 490                         addr = eva;
 491                         break;
 492
 493                 case T_DIVIDE:          /* integer divide fault */
 494                         ucode = FPE_INTDIV;
 495                         i = SIGFPE;
 496                         break;
 497
 498 #ifdef DEV_ISA
 499                 case T_NMI:
 500 #ifdef POWERFAIL_NMI
 501 #ifndef TIMER_FREQ
 502 #  define TIMER_FREQ 1193182
 503 #endif
 504                         if (time_second - lastalert > 10) {
 505                                 log(LOG_WARNING, "NMI: power fail\n");
 506                                 sysbeep(880, hz);
 507                                 lastalert = time_second;
 508                         }
 509                         goto userout;
 510 #else /* !POWERFAIL_NMI */
 511                         /* machine/parity/power fail/"kitchen sink" faults */
 512                         if (isa_nmi(code) == 0) {
 513 #ifdef KDB
 514                                 /*
 515                                  * NMI can be hooked up to a pushbutton
 516                                  * for debugging.
 517                                  */
 518                                 if (kdb_on_nmi) {
 519                                         printf ("NMI ... going to debugger\n");
 520                                         kdb_trap(type, 0, frame);
 521                                 }
 522 #endif /* KDB */
 523                                 goto userout;
 524                         } else if (panic_on_nmi)
 525                                 panic("NMI indicates hardware failure");
 526                         break;
 527 #endif /* POWERFAIL_NMI */
 528 #endif /* DEV_ISA */
 529
 530                 case T_OFLOW:           /* integer overflow fault */
 531                         ucode = FPE_INTOVF;
 532                         i = SIGFPE;
 533                         break;
 534
 535                 case T_BOUND:           /* bounds check fault */
 536                         ucode = FPE_FLTSUB;
 537                         i = SIGFPE;
 538                         break;
 539
 540                 case T_DNA:
 541 #ifdef DEV_NPX
 542                         KASSERT(PCB_USER_FPU(td->td_pcb),
 543                             ("kernel FPU ctx has leaked"));
 544                         /* transparent fault (due to context switch "late") */
 545                         if (npxdna())
 546                                 goto userout;
 547 #endif
 548                         uprintf("pid %d killed due to lack of floating point\n",
 549                                 p->p_pid);
 550                         i = SIGKILL;
 551                         ucode = 0;
 552                         break;
 553
 554                 case T_FPOPFLT:         /* FPU operand fetch fault */
 555                         ucode = ILL_COPROC;
 556                         i = SIGILL;
 557                         break;
 558
 559                 case T_XMMFLT:          /* SIMD floating-point exception */
 560                         ucode = 0; /* XXX */
 561                         i = SIGFPE;
 562                         break;
 563                 }
 564         } else {
 565                 /* kernel trap */
 566
 567                 KASSERT(cold || td->td_ucred != NULL,
 568                     ("kernel trap doesn't have ucred"));
 569                 switch (type) {
 570                 case T_PAGEFLT:                 /* page fault */
 571                         (void) trap_pfault(frame, FALSE, eva);
 572                         goto out;
 573
 574                 case T_DNA:
 575 #ifdef DEV_NPX
 576                         KASSERT(!PCB_USER_FPU(td->td_pcb),
 577                             ("Unregistered use of FPU in kernel"));
 578                         if (npxdna())
 579                                 goto out;
 580 #endif
 581                         break;
 582
 583                 case T_ARITHTRAP:       /* arithmetic trap */
 584                 case T_XMMFLT:          /* SIMD floating-point exception */
 585                 case T_FPOPFLT:         /* FPU operand fetch fault */
 586                         /*
 587                          * XXXKIB for now disable any FPU traps in kernel
 588                          * handler registration seems to be overkill
 589                          */
 590                         trap_fatal(frame, 0);
 591                         goto out;
 592
 593                         /*
 594                          * The following two traps can happen in
 595                          * vm86 mode, and, if so, we want to handle
 596                          * them specially.
 597                          */
 598                 case T_PROTFLT:         /* general protection fault */
 599                 case T_STKFLT:          /* stack fault */
 600                         if (frame->tf_eflags & PSL_VM) {
 601                                 i = vm86_emulate((struct vm86frame *)frame);
 602                                 if (i != 0)
 603                                         /*
 604                                          * returns to original process
 605                                          */
 606                                         vm86_trap((struct vm86frame *)frame);
 607                                 goto out;
 608                         }
 609                         if (type == T_STKFLT)
 610                                 break;
 611
 612                         /* FALL THROUGH */
 613
 614                 case T_SEGNPFLT:        /* segment not present fault */
 615                         if (PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL)
 616                                 break;
 617
 618                         /*
 619                          * Invalid %fs's and %gs's can be created using
 620                          * procfs or PT_SETREGS or by invalidating the
 621                          * underlying LDT entry.  This causes a fault
 622                          * in kernel mode when the kernel attempts to
 623                          * switch contexts.  Lose the bad context
 624                          * (XXX) so that we can continue, and generate
 625                          * a signal.
 626                          */
 627                         if (frame->tf_eip == (int)cpu_switch_load_gs) {
 628                                 PCPU_GET(curpcb)->pcb_gs = 0;
 629 #if 0
 630                                 PROC_LOCK(p);
 631                                 psignal(p, SIGBUS);
 632                                 PROC_UNLOCK(p);
 633 #endif
 634                                 goto out;
 635                         }
 636
 637                         if (td->td_intr_nesting_level != 0)
 638                                 break;
 639
 640                         /*
 641                          * Invalid segment selectors and out of bounds
 642                          * %eip's and %esp's can be set up in user mode.
 643                          * This causes a fault in kernel mode when the
 644                          * kernel tries to return to user mode.  We want
 645                          * to get this fault so that we can fix the
 646                          * problem here and not have to check all the
 647                          * selectors and pointers when the user changes
 648                          * them.
 649                          */
 650                         if (frame->tf_eip == (int)doreti_iret) {
 651                                 frame->tf_eip = (int)doreti_iret_fault;
 652                                 goto out;
 653                         }
 654                         if (frame->tf_eip == (int)doreti_popl_ds) {
 655                                 frame->tf_eip = (int)doreti_popl_ds_fault;
 656                                 goto out;
 657                         }
 658                         if (frame->tf_eip == (int)doreti_popl_es) {
 659                                 frame->tf_eip = (int)doreti_popl_es_fault;
 660                                 goto out;
 661                         }
 662                         if (frame->tf_eip == (int)doreti_popl_fs) {
 663                                 frame->tf_eip = (int)doreti_popl_fs_fault;
 664                                 goto out;
 665                         }
 666                         if (PCPU_GET(curpcb)->pcb_onfault != NULL) {
 667                                 frame->tf_eip =
 668                                     (int)PCPU_GET(curpcb)->pcb_onfault;
 669                                 goto out;
 670                         }
 671                         break;
 672
 673                 case T_TSSFLT:
 674                         /*
 675                          * PSL_NT can be set in user mode and isn't cleared
 676                          * automatically when the kernel is entered.  This
 677                          * causes a TSS fault when the kernel attempts to
 678                          * `iret' because the TSS link is uninitialized.  We
 679                          * want to get this fault so that we can fix the
 680                          * problem here and not every time the kernel is
 681                          * entered.
 682                          */
 683                         if (frame->tf_eflags & PSL_NT) {
 684                                 frame->tf_eflags &= ~PSL_NT;
 685                                 goto out;
 686                         }
 687                         break;
 688
 689                 case T_TRCTRAP:  /* trace trap */
 690                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall)) {
 691                                 /*
 692                                  * We've just entered system mode via the
 693                                  * syscall lcall.  Continue single stepping
 694                                  * silently until the syscall handler has
 695                                  * saved the flags.
 696                                  */
 697                                 goto out;
 698                         }
 699                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall) + 1) {
 700                                 /*
 701                                  * The syscall handler has now saved the
 702                                  * flags.  Stop single stepping it.
 703                                  */
 704                                 frame->tf_eflags &= ~PSL_T;
 705                                 goto out;
 706                         }
 707                         /*
 708                          * Ignore debug register trace traps due to
 709                          * accesses in the user's address space, which
 710                          * can happen under several conditions such as
 711                          * if a user sets a watchpoint on a buffer and
 712                          * then passes that buffer to a system call.
 713                          * We still want to get TRCTRAPS for addresses
 714                          * in kernel space because that is useful when
 715                          * debugging the kernel.
 716                          */
 717                         if (user_dbreg_trap() &&
 718                            !(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL)) {
 719                                 /*
 720                                  * Reset breakpoint bits because the
 721                                  * processor doesn't
 722                                  */
 723                                 load_dr6(rdr6() & 0xfffffff0);
 724                                 goto out;
 725                         }
 726                         /*
 727                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 728                          */
 729                 case T_BPTFLT:
 730                         /*
 731                          * If KDB is enabled, let it handle the debugger trap.
 732                          * Otherwise, debugger traps "can't happen".
 733                          */
 734 #ifdef KDB
 735                         if (kdb_trap(type, 0, frame))
 736                                 goto out;
 737 #endif
 738                         break;
 739
 740 #ifdef DEV_ISA
 741                 case T_NMI:
 742 #ifdef POWERFAIL_NMI
 743                         if (time_second - lastalert > 10) {
 744                                 log(LOG_WARNING, "NMI: power fail\n");
 745                                 sysbeep(880, hz);
 746                                 lastalert = time_second;
 747                         }
 748                         goto out;
 749 #else /* !POWERFAIL_NMI */
 750                         /* machine/parity/power fail/"kitchen sink" faults */
 751                         if (isa_nmi(code) == 0) {
 752 #ifdef KDB
 753                                 /*
 754                                  * NMI can be hooked up to a pushbutton
 755                                  * for debugging.
 756                                  */
 757                                 if (kdb_on_nmi) {
 758                                         printf ("NMI ... going to debugger\n");
 759                                         kdb_trap(type, 0, frame);
 760                                 }
 761 #endif /* KDB */
 762                                 goto out;
 763                         } else if (panic_on_nmi == 0)
 764                                 goto out;
 765                         /* FALLTHROUGH */
 766 #endif /* POWERFAIL_NMI */
 767 #endif /* DEV_ISA */
 768                 }
 769
 770                 trap_fatal(frame, eva);
 771                 goto out;
 772         }
 773
 774         /* Translate fault for emulators (e.g. Linux) */
 775         if (*p->p_sysent->sv_transtrap)
 776                 i = (*p->p_sysent->sv_transtrap)(i, type);
 777
 778         ksiginfo_init_trap(&ksi);
 779         ksi.ksi_signo = i;
 780         ksi.ksi_code = ucode;
 781         ksi.ksi_addr = (void *)addr;
 782         ksi.ksi_trapno = type;
 783         trapsignal(td, &ksi);
 784
 785 #ifdef DEBUG
 786         if (type <= MAX_TRAP_MSG) {
 787                 uprintf("fatal process exception: %s",
 788                         trap_msg[type]);
 789                 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
 790                         uprintf(", fault VA = 0x%lx", (u_long)eva);
 791                 uprintf("\n");
 792         }
 793 #endif
 794
 795 user:
 796         userret(td, frame);
 797         mtx_assert(&Giant, MA_NOTOWNED);
 798         KASSERT(PCB_USER_FPU(td->td_pcb),
 799             ("Return from trap with kernel FPU ctx leaked"));
 800 userout:
 801 out:
 802         return;
 803 }
 804
 805 static int
 806 trap_pfault(frame, usermode, eva)
 807         struct trapframe *frame;
 808         int usermode;
 809         vm_offset_t eva;
 810 {
 811         vm_offset_t va;
 812         struct vmspace *vm = NULL;
 813         vm_map_t map;
 814         int rv = 0;
 815         vm_prot_t ftype;
 816         struct thread *td = curthread;
 817         struct proc *p = td->td_proc;
 818
 819         va = trunc_page(eva);
 820         if (va >= KERNBASE) {
 821                 /*
 822                  * Don't allow user-mode faults in kernel address space.
 823                  * An exception:  if the faulting address is the invalid
 824                  * instruction entry in the IDT, then the Intel Pentium
 825                  * F00F bug workaround was triggered, and we need to
 826                  * treat it is as an illegal instruction, and not a page
 827                  * fault.
 828                  */
 829 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 830                 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
 831                         return -2;
 832 #endif
 833                 if (usermode)
 834                         goto nogo;
 835
 836                 map = kernel_map;
 837         } else {
 838                 /*
 839                  * This is a fault on non-kernel virtual memory.
 840                  * vm is initialized above to NULL. If curproc is NULL
 841                  * or curproc->p_vmspace is NULL the fault is fatal.
 842                  */
 843                 if (p != NULL)
 844                         vm = p->p_vmspace;
 845
 846                 if (vm == NULL)
 847                         goto nogo;
 848
 849                 map = &vm->vm_map;
 850         }
 851
 852         /*
 853          * PGEX_I is defined only if the execute disable bit capability is
 854          * supported and enabled.
 855          */
 856         if (frame->tf_err & PGEX_W)
 857                 ftype = VM_PROT_WRITE;
 858 #ifdef PAE
 859         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 860                 ftype = VM_PROT_EXECUTE;
 861 #endif
 862         else
 863                 ftype = VM_PROT_READ;
 864
 865         if (map != kernel_map) {
 866                 /*
 867                  * Keep swapout from messing with us during this
 868                  *      critical time.
 869                  */
 870                 PROC_LOCK(p);
 871                 ++p->p_lock;
 872                 PROC_UNLOCK(p);
 873
 874                 /* Fault in the user page: */
 875                 rv = vm_fault(map, va, ftype,
 876                               (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
 877                                                       : VM_FAULT_NORMAL);
 878
 879                 PROC_LOCK(p);
 880                 --p->p_lock;
 881                 PROC_UNLOCK(p);
 882         } else {
 883                 /*
 884                  * Don't have to worry about process locking or stacks in the
 885                  * kernel.
 886                  */
 887                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 888         }
 889         if (rv == KERN_SUCCESS)
 890                 return (0);
 891 nogo:
 892         if (!usermode) {
 893                 if (td->td_intr_nesting_level == 0 &&
 894                     PCPU_GET(curpcb)->pcb_onfault != NULL) {
 895                         frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault;
 896                         return (0);
 897                 }
 898                 trap_fatal(frame, eva);
 899                 return (-1);
 900         }
 901
 902         return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 903 }
 904
 905 static void
 906 trap_fatal(frame, eva)
 907         struct trapframe *frame;
 908         vm_offset_t eva;
 909 {
 910         int code, ss, esp;
 911         u_int type;
 912         struct soft_segment_descriptor softseg;
 913         char *msg;
 914
 915         code = frame->tf_err;
 916         type = frame->tf_trapno;
 917         sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
 918
 919         if (type <= MAX_TRAP_MSG)
 920                 msg = trap_msg[type];
 921         else
 922                 msg = "UNKNOWN";
 923         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 924             frame->tf_eflags & PSL_VM ? "vm86" :
 925             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
 926 #ifdef SMP
 927         /* two separate prints in case of a trap on an unmapped page */
 928         printf("cpuid = %d; ", PCPU_GET(cpuid));
 929         printf("apic id = %02x\n", PCPU_GET(apic_id));
 930 #endif
 931         if (type == T_PAGEFLT) {
 932                 printf("fault virtual address   = 0x%x\n", eva);
 933                 printf("fault code              = %s %s, %s\n",
 934                         code & PGEX_U ? "user" : "supervisor",
 935                         code & PGEX_W ? "write" : "read",
 936                         code & PGEX_P ? "protection violation" : "page not present");
 937         }
 938         printf("instruction pointer     = 0x%x:0x%x\n",
 939                frame->tf_cs & 0xffff, frame->tf_eip);
 940         if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
 941                 ss = frame->tf_ss & 0xffff;
 942                 esp = frame->tf_esp;
 943         } else {
 944                 ss = GSEL(GDATA_SEL, SEL_KPL);
 945                 esp = (int)&frame->tf_esp;
 946         }
 947         printf("stack pointer           = 0x%x:0x%x\n", ss, esp);
 948         printf("frame pointer           = 0x%x:0x%x\n", ss, frame->tf_ebp);
 949         printf("code segment            = base 0x%x, limit 0x%x, type 0x%x\n",
 950                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 951         printf("                        = DPL %d, pres %d, def32 %d, gran %d\n",
 952                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
 953                softseg.ssd_gran);
 954         printf("processor eflags        = ");
 955         if (frame->tf_eflags & PSL_T)
 956                 printf("trace trap, ");
 957         if (frame->tf_eflags & PSL_I)
 958                 printf("interrupt enabled, ");
 959         if (frame->tf_eflags & PSL_NT)
 960                 printf("nested task, ");
 961         if (frame->tf_eflags & PSL_RF)
 962                 printf("resume, ");
 963         if (frame->tf_eflags & PSL_VM)
 964                 printf("vm86, ");
 965         printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
 966         printf("current process         = ");
 967         if (curproc) {
 968                 printf("%lu (%s)\n", (u_long)curproc->p_pid, curthread->td_name);
 969         } else {
 970                 printf("Idle\n");
 971         }
 972
 973 #ifdef KDB
 974         if (debugger_on_panic || kdb_active) {
 975                 frame->tf_err = eva;    /* smuggle fault address to ddb */
 976                 if (kdb_trap(type, 0, frame)) {
 977                         frame->tf_err = code;   /* restore error code */
 978                         return;
 979                 }
 980                 frame->tf_err = code;           /* restore error code */
 981         }
 982 #endif
 983         printf("trap number             = %d\n", type);
 984         if (type <= MAX_TRAP_MSG)
 985                 panic("%s", trap_msg[type]);
 986         else
 987                 panic("unknown/reserved trap");
 988 }
 989
 990 /*
 991  * Double fault handler. Called when a fault occurs while writing
 992  * a frame for a trap/exception onto the stack. This usually occurs
 993  * when the stack overflows (such is the case with infinite recursion,
 994  * for example).
 995  *
 996  * XXX Note that the current PTD gets replaced by IdlePTD when the
 997  * task switch occurs. This means that the stack that was active at
 998  * the time of the double fault is not available at <kstack> unless
 999  * the machine was idle when the double fault occurred. The downside
1000  * of this is that "trace <ebp>" in ddb won't work.
1001  */
1002 void
1003 dblfault_handler()
1004 {
1005 #ifdef KDTRACE_HOOKS
1006         if (dtrace_doubletrap_func != NULL)
1007                 (*dtrace_doubletrap_func)();
1008 #endif
1009         printf("\nFatal double fault:\n");
1010         printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip));
1011         printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp));
1012         printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp));
1013 #ifdef SMP
1014         /* two separate prints in case of a trap on an unmapped page */
1015         printf("cpuid = %d; ", PCPU_GET(cpuid));
1016         printf("apic id = %02x\n", PCPU_GET(apic_id));
1017 #endif
1018         panic("double fault");
1019 }
1020
1021 int
1022 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
1023 {
1024         struct proc *p;
1025         struct trapframe *frame;
1026         caddr_t params;
1027         int error;
1028
1029         p = td->td_proc;
1030         frame = td->td_frame;
1031
1032         params = (caddr_t)frame->tf_esp + sizeof(int);
1033         sa->code = frame->tf_eax;
1034
1035         /*
1036          * Need to check if this is a 32 bit or 64 bit syscall.
1037          */
1038         if (sa->code == SYS_syscall) {
1039                 /*
1040                  * Code is first argument, followed by actual args.
1041                  */
1042                 sa->code = fuword(params);
1043                 params += sizeof(int);
1044         } else if (sa->code == SYS___syscall) {
1045                 /*
1046                  * Like syscall, but code is a quad, so as to maintain
1047                  * quad alignment for the rest of the arguments.
1048                  */
1049                 sa->code = fuword(params);
1050                 params += sizeof(quad_t);
1051         }
1052
1053         if (p->p_sysent->sv_mask)
1054                 sa->code &= p->p_sysent->sv_mask;
1055         if (sa->code >= p->p_sysent->sv_size)
1056                 sa->callp = &p->p_sysent->sv_table[0];
1057         else
1058                 sa->callp = &p->p_sysent->sv_table[sa->code];
1059         sa->narg = sa->callp->sy_narg;
1060
1061         if (params != NULL && sa->narg != 0)
1062                 error = copyin(params, (caddr_t)sa->args,
1063                     (u_int)(sa->narg * sizeof(int)));
1064         else
1065                 error = 0;
1066
1067         if (error == 0) {
1068                 td->td_retval[0] = 0;
1069                 td->td_retval[1] = frame->tf_edx;
1070         }
1071
1072         return (error);
1073 }
1074
1075 /*
1076  *      syscall -       system call request C handler
1077  *
1078  *      A system call is essentially treated as a trap.
1079  */
1080 void
1081 syscall(struct trapframe *frame)
1082 {
1083         struct thread *td;
1084         struct syscall_args sa;
1085         register_t orig_tf_eflags;
1086         int error;
1087         ksiginfo_t ksi;
1088
1089 #ifdef DIAGNOSTIC
1090         if (ISPL(frame->tf_cs) != SEL_UPL) {
1091                 panic("syscall");
1092                 /* NOT REACHED */
1093         }
1094 #endif
1095         orig_tf_eflags = frame->tf_eflags;
1096
1097         td = curthread;
1098         td->td_frame = frame;
1099
1100         error = syscallenter(td, &sa);
1101
1102         /*
1103          * Traced syscall.
1104          */
1105         if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1106                 frame->tf_eflags &= ~PSL_T;
1107                 ksiginfo_init_trap(&ksi);
1108                 ksi.ksi_signo = SIGTRAP;
1109                 ksi.ksi_code = TRAP_TRACE;
1110                 ksi.ksi_addr = (void *)frame->tf_eip;
1111                 trapsignal(td, &ksi);
1112         }
1113
1114         KASSERT(PCB_USER_FPU(td->td_pcb),
1115             ("System call %s returning with kernel FPU ctx leaked",
1116              syscallname(td->td_proc, sa.code)));
1117         KASSERT(td->td_pcb->pcb_save == &td->td_pcb->pcb_user_save,
1118             ("System call %s returning with mangled pcb_save",
1119              syscallname(td->td_proc, sa.code)));
1120
1121         syscallret(td, error, &sa);
1122 }