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_npx.h"
  53 #include "opt_stack.h"
  54 #include "opt_trap.h"
  55
  56 #include <sys/param.h>
  57 #include <sys/bus.h>
  58 #include <sys/systm.h>
  59 #include <sys/proc.h>
  60 #include <sys/pioctl.h>
  61 #include <sys/ptrace.h>
  62 #include <sys/kdb.h>
  63 #include <sys/kernel.h>
  64 #include <sys/ktr.h>
  65 #include <sys/lock.h>
  66 #include <sys/mutex.h>
  67 #include <sys/resourcevar.h>
  68 #include <sys/signalvar.h>
  69 #include <sys/syscall.h>
  70 #include <sys/sysctl.h>
  71 #include <sys/sysent.h>
  72 #include <sys/uio.h>
  73 #include <sys/vmmeter.h>
  74 #ifdef HWPMC_HOOKS
  75 #include <sys/pmckern.h>
  76 PMC_SOFT_DEFINE( , , page_fault, all);
  77 PMC_SOFT_DEFINE( , , page_fault, read);
  78 PMC_SOFT_DEFINE( , , page_fault, write);
  79 #endif
  80 #include <security/audit/audit.h>
  81
  82 #include <vm/vm.h>
  83 #include <vm/vm_param.h>
  84 #include <vm/pmap.h>
  85 #include <vm/vm_kern.h>
  86 #include <vm/vm_map.h>
  87 #include <vm/vm_page.h>
  88 #include <vm/vm_extern.h>
  89
  90 #include <machine/cpu.h>
  91 #include <machine/intr_machdep.h>
  92 #include <x86/mca.h>
  93 #include <machine/md_var.h>
  94 #include <machine/pcb.h>
  95 #ifdef SMP
  96 #include <machine/smp.h>
  97 #endif
  98 #include <machine/stack.h>
  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 #endif
 110
 111 extern void trap(struct trapframe *frame);
 112 extern void syscall(struct trapframe *frame);
 113
 114 static int trap_pfault(struct trapframe *, int, vm_offset_t);
 115 static void trap_fatal(struct trapframe *, vm_offset_t);
 116 void dblfault_handler(void);
 117
 118 extern inthand_t IDTVEC(lcall_syscall);
 119
 120 #define MAX_TRAP_MSG            32
 121 static char *trap_msg[] = {
 122         "",                                     /*  0 unused */
 123         "privileged instruction fault",         /*  1 T_PRIVINFLT */
 124         "",                                     /*  2 unused */
 125         "breakpoint instruction fault",         /*  3 T_BPTFLT */
 126         "",                                     /*  4 unused */
 127         "",                                     /*  5 unused */
 128         "arithmetic trap",                      /*  6 T_ARITHTRAP */
 129         "",                                     /*  7 unused */
 130         "",                                     /*  8 unused */
 131         "general protection fault",             /*  9 T_PROTFLT */
 132         "trace trap",                           /* 10 T_TRCTRAP */
 133         "",                                     /* 11 unused */
 134         "page fault",                           /* 12 T_PAGEFLT */
 135         "",                                     /* 13 unused */
 136         "alignment fault",                      /* 14 T_ALIGNFLT */
 137         "",                                     /* 15 unused */
 138         "",                                     /* 16 unused */
 139         "",                                     /* 17 unused */
 140         "integer divide fault",                 /* 18 T_DIVIDE */
 141         "non-maskable interrupt trap",          /* 19 T_NMI */
 142         "overflow trap",                        /* 20 T_OFLOW */
 143         "FPU bounds check fault",               /* 21 T_BOUND */
 144         "FPU device not available",             /* 22 T_DNA */
 145         "double fault",                         /* 23 T_DOUBLEFLT */
 146         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
 147         "invalid TSS fault",                    /* 25 T_TSSFLT */
 148         "segment not present fault",            /* 26 T_SEGNPFLT */
 149         "stack fault",                          /* 27 T_STKFLT */
 150         "machine check trap",                   /* 28 T_MCHK */
 151         "SIMD floating-point exception",        /* 29 T_XMMFLT */
 152         "reserved (unknown) fault",             /* 30 T_RESERVED */
 153         "",                                     /* 31 unused (reserved) */
 154         "DTrace pid return trap",               /* 32 T_DTRACE_RET */
 155 };
 156
 157 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 158 int has_f00f_bug = 0;           /* Initialized so that it can be patched. */
 159 #endif
 160
 161 #ifdef KDB
 162 static int kdb_on_nmi = 1;
 163 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RWTUN,
 164         &kdb_on_nmi, 0, "Go to KDB on NMI");
 165 #endif
 166 static int panic_on_nmi = 1;
 167 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RWTUN,
 168         &panic_on_nmi, 0, "Panic on NMI");
 169 static int prot_fault_translation = 0;
 170 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
 171         &prot_fault_translation, 0, "Select signal to deliver on protection fault");
 172 static int uprintf_signal;
 173 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW,
 174     &uprintf_signal, 0,
 175     "Print debugging information on trap signal to ctty");
 176
 177 /*
 178  * Exception, fault, and trap interface to the FreeBSD kernel.
 179  * This common code is called from assembly language IDT gate entry
 180  * routines that prepare a suitable stack frame, and restore this
 181  * frame after the exception has been processed.
 182  */
 183
 184 void
 185 trap(struct trapframe *frame)
 186 {
 187 #ifdef KDTRACE_HOOKS
 188         struct reg regs;
 189 #endif
 190         struct thread *td = curthread;
 191         struct proc *p = td->td_proc;
 192 #ifdef KDB
 193         register_t dr6;
 194 #endif
 195         int i = 0, ucode = 0;
 196         u_int type;
 197         register_t addr = 0;
 198         vm_offset_t eva;
 199         ksiginfo_t ksi;
 200 #ifdef POWERFAIL_NMI
 201         static int lastalert = 0;
 202 #endif
 203
 204         PCPU_INC(cnt.v_trap);
 205         type = frame->tf_trapno;
 206
 207 #ifdef SMP
 208         /* Handler for NMI IPIs used for stopping CPUs. */
 209         if (type == T_NMI) {
 210                  if (ipi_nmi_handler() == 0)
 211                            goto out;
 212         }
 213 #endif /* SMP */
 214
 215 #ifdef KDB
 216         if (kdb_active) {
 217                 kdb_reenter();
 218                 goto out;
 219         }
 220 #endif
 221
 222         if (type == T_RESERVED) {
 223                 trap_fatal(frame, 0);
 224                 goto out;
 225         }
 226
 227         if (type == T_NMI) {
 228 #ifdef HWPMC_HOOKS
 229                 /*
 230                  * CPU PMCs interrupt using an NMI so we check for that first.
 231                  * If the HWPMC module is active, 'pmc_hook' will point to
 232                  * the function to be called.  A non-zero return value from the
 233                  * hook means that the NMI was consumed by it and that we can
 234                  * return immediately.
 235                  */
 236                 if (pmc_intr != NULL &&
 237                     (*pmc_intr)(PCPU_GET(cpuid), frame) != 0)
 238                         goto out;
 239 #endif
 240
 241 #ifdef STACK
 242                 if (stack_nmi_handler(frame) != 0)
 243                         goto out;
 244 #endif
 245         }
 246
 247         if (type == T_MCHK) {
 248                 mca_intr();
 249                 goto out;
 250         }
 251
 252 #ifdef KDTRACE_HOOKS
 253         /*
 254          * A trap can occur while DTrace executes a probe. Before
 255          * executing the probe, DTrace blocks re-scheduling and sets
 256          * a flag in its per-cpu flags to indicate that it doesn't
 257          * want to fault. On returning from the probe, the no-fault
 258          * flag is cleared and finally re-scheduling is enabled.
 259          */
 260         if ((type == T_PROTFLT || type == T_PAGEFLT) &&
 261             dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
 262                 goto out;
 263 #endif
 264
 265         if ((frame->tf_eflags & PSL_I) == 0) {
 266                 /*
 267                  * Buggy application or kernel code has disabled
 268                  * interrupts and then trapped.  Enabling interrupts
 269                  * now is wrong, but it is better than running with
 270                  * interrupts disabled until they are accidentally
 271                  * enabled later.
 272                  */
 273                 if (TRAPF_USERMODE(frame) &&
 274                     (curpcb->pcb_flags & PCB_VM86CALL) == 0)
 275                         uprintf(
 276                             "pid %ld (%s): trap %d with interrupts disabled\n",
 277                             (long)curproc->p_pid, curthread->td_name, type);
 278                 else if (type != T_NMI && type != T_BPTFLT &&
 279                     type != T_TRCTRAP &&
 280                     frame->tf_eip != (int)cpu_switch_load_gs) {
 281                         /*
 282                          * XXX not quite right, since this may be for a
 283                          * multiple fault in user mode.
 284                          */
 285                         printf("kernel trap %d with interrupts disabled\n",
 286                             type);
 287                         /*
 288                          * Page faults need interrupts disabled until later,
 289                          * and we shouldn't enable interrupts while holding
 290                          * a spin lock.
 291                          */
 292                         if (type != T_PAGEFLT &&
 293                             td->td_md.md_spinlock_count == 0)
 294                                 enable_intr();
 295                 }
 296         }
 297         eva = 0;
 298         if (type == T_PAGEFLT) {
 299                 /*
 300                  * For some Cyrix CPUs, %cr2 is clobbered by
 301                  * interrupts.  This problem is worked around by using
 302                  * an interrupt gate for the pagefault handler.  We
 303                  * are finally ready to read %cr2 and conditionally
 304                  * reenable interrupts.  If we hold a spin lock, then
 305                  * we must not reenable interrupts.  This might be a
 306                  * spurious page fault.
 307                  */
 308                 eva = rcr2();
 309                 if (td->td_md.md_spinlock_count == 0)
 310                         enable_intr();
 311         }
 312
 313         if (TRAPF_USERMODE(frame) && (curpcb->pcb_flags & PCB_VM86CALL) == 0) {
 314                 /* user trap */
 315
 316                 td->td_pticks = 0;
 317                 td->td_frame = frame;
 318                 addr = frame->tf_eip;
 319                 if (td->td_cowgen != p->p_cowgen)
 320                         thread_cow_update(td);
 321
 322                 switch (type) {
 323                 case T_PRIVINFLT:       /* privileged instruction fault */
 324                         i = SIGILL;
 325                         ucode = ILL_PRVOPC;
 326                         break;
 327
 328                 case T_BPTFLT:          /* bpt instruction fault */
 329                 case T_TRCTRAP:         /* trace trap */
 330                         enable_intr();
 331 #ifdef KDTRACE_HOOKS
 332                         if (type == T_BPTFLT) {
 333                                 fill_frame_regs(frame, &regs);
 334                                 if (dtrace_pid_probe_ptr != NULL &&
 335                                     dtrace_pid_probe_ptr(&regs) == 0)
 336                                         goto out;
 337                         }
 338 #endif
 339 user_trctrap_out:
 340                         frame->tf_eflags &= ~PSL_T;
 341                         i = SIGTRAP;
 342                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 343                         break;
 344
 345                 case T_ARITHTRAP:       /* arithmetic trap */
 346 #ifdef DEV_NPX
 347                         ucode = npxtrap_x87();
 348                         if (ucode == -1)
 349                                 goto userout;
 350 #else
 351                         ucode = 0;
 352 #endif
 353                         i = SIGFPE;
 354                         break;
 355
 356                         /*
 357                          * The following two traps can happen in
 358                          * vm86 mode, and, if so, we want to handle
 359                          * them specially.
 360                          */
 361                 case T_PROTFLT:         /* general protection fault */
 362                 case T_STKFLT:          /* stack fault */
 363                         if (frame->tf_eflags & PSL_VM) {
 364                                 i = vm86_emulate((struct vm86frame *)frame);
 365                                 if (i == SIGTRAP) {
 366                                         type = T_TRCTRAP;
 367                                         load_dr6(rdr6() | 0x4000);
 368                                         goto user_trctrap_out;
 369                                 }
 370                                 if (i == 0)
 371                                         goto user;
 372                                 break;
 373                         }
 374                         i = SIGBUS;
 375                         ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
 376                         break;
 377                 case T_SEGNPFLT:        /* segment not present fault */
 378                         i = SIGBUS;
 379                         ucode = BUS_ADRERR;
 380                         break;
 381                 case T_TSSFLT:          /* invalid TSS fault */
 382                         i = SIGBUS;
 383                         ucode = BUS_OBJERR;
 384                         break;
 385                 case T_ALIGNFLT:
 386                         i = SIGBUS;
 387                         ucode = BUS_ADRALN;
 388                         break;
 389                 case T_DOUBLEFLT:       /* double fault */
 390                 default:
 391                         i = SIGBUS;
 392                         ucode = BUS_OBJERR;
 393                         break;
 394
 395                 case T_PAGEFLT:         /* page fault */
 396
 397                         i = trap_pfault(frame, TRUE, eva);
 398 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 399                         if (i == -2) {
 400                                 /*
 401                                  * The f00f hack workaround has triggered, so
 402                                  * treat the fault as an illegal instruction
 403                                  * (T_PRIVINFLT) instead of a page fault.
 404                                  */
 405                                 type = frame->tf_trapno = T_PRIVINFLT;
 406
 407                                 /* Proceed as in that case. */
 408                                 ucode = ILL_PRVOPC;
 409                                 i = SIGILL;
 410                                 break;
 411                         }
 412 #endif
 413                         if (i == -1)
 414                                 goto userout;
 415                         if (i == 0)
 416                                 goto user;
 417
 418                         if (i == SIGSEGV)
 419                                 ucode = SEGV_MAPERR;
 420                         else {
 421                                 if (prot_fault_translation == 0) {
 422                                         /*
 423                                          * Autodetect.
 424                                          * This check also covers the images
 425                                          * without the ABI-tag ELF note.
 426                                          */
 427                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 428                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 429                                                 i = SIGSEGV;
 430                                                 ucode = SEGV_ACCERR;
 431                                         } else {
 432                                                 i = SIGBUS;
 433                                                 ucode = BUS_PAGE_FAULT;
 434                                         }
 435                                 } else if (prot_fault_translation == 1) {
 436                                         /*
 437                                          * Always compat mode.
 438                                          */
 439                                         i = SIGBUS;
 440                                         ucode = BUS_PAGE_FAULT;
 441                                 } else {
 442                                         /*
 443                                          * Always SIGSEGV mode.
 444                                          */
 445                                         i = SIGSEGV;
 446                                         ucode = SEGV_ACCERR;
 447                                 }
 448                         }
 449                         addr = eva;
 450                         break;
 451
 452                 case T_DIVIDE:          /* integer divide fault */
 453                         ucode = FPE_INTDIV;
 454                         i = SIGFPE;
 455                         break;
 456
 457 #ifdef DEV_ISA
 458                 case T_NMI:
 459 #ifdef POWERFAIL_NMI
 460 #ifndef TIMER_FREQ
 461 #  define TIMER_FREQ 1193182
 462 #endif
 463                         if (time_second - lastalert > 10) {
 464                                 log(LOG_WARNING, "NMI: power fail\n");
 465                                 sysbeep(880, hz);
 466                                 lastalert = time_second;
 467                         }
 468                         goto userout;
 469 #else /* !POWERFAIL_NMI */
 470                         /* machine/parity/power fail/"kitchen sink" faults */
 471                         if (isa_nmi(frame->tf_err) == 0) {
 472 #ifdef KDB
 473                                 /*
 474                                  * NMI can be hooked up to a pushbutton
 475                                  * for debugging.
 476                                  */
 477                                 if (kdb_on_nmi) {
 478                                         printf ("NMI ... going to debugger\n");
 479                                         kdb_trap(type, 0, frame);
 480                                 }
 481 #endif /* KDB */
 482                                 goto userout;
 483                         } else if (panic_on_nmi)
 484                                 panic("NMI indicates hardware failure");
 485                         break;
 486 #endif /* POWERFAIL_NMI */
 487 #endif /* DEV_ISA */
 488
 489                 case T_OFLOW:           /* integer overflow fault */
 490                         ucode = FPE_INTOVF;
 491                         i = SIGFPE;
 492                         break;
 493
 494                 case T_BOUND:           /* bounds check fault */
 495                         ucode = FPE_FLTSUB;
 496                         i = SIGFPE;
 497                         break;
 498
 499                 case T_DNA:
 500 #ifdef DEV_NPX
 501                         KASSERT(PCB_USER_FPU(td->td_pcb),
 502                             ("kernel FPU ctx has leaked"));
 503                         /* transparent fault (due to context switch "late") */
 504                         if (npxdna())
 505                                 goto userout;
 506 #endif
 507                         uprintf("pid %d killed due to lack of floating point\n",
 508                                 p->p_pid);
 509                         i = SIGKILL;
 510                         ucode = 0;
 511                         break;
 512
 513                 case T_FPOPFLT:         /* FPU operand fetch fault */
 514                         ucode = ILL_COPROC;
 515                         i = SIGILL;
 516                         break;
 517
 518                 case T_XMMFLT:          /* SIMD floating-point exception */
 519 #if defined(DEV_NPX) && !defined(CPU_DISABLE_SSE) && defined(I686_CPU)
 520                         ucode = npxtrap_sse();
 521                         if (ucode == -1)
 522                                 goto userout;
 523 #else
 524                         ucode = 0;
 525 #endif
 526                         i = SIGFPE;
 527                         break;
 528 #ifdef KDTRACE_HOOKS
 529                 case T_DTRACE_RET:
 530                         enable_intr();
 531                         fill_frame_regs(frame, &regs);
 532                         if (dtrace_return_probe_ptr != NULL &&
 533                             dtrace_return_probe_ptr(&regs) == 0)
 534                                 goto out;
 535                         break;
 536 #endif
 537                 }
 538         } else {
 539                 /* kernel trap */
 540
 541                 KASSERT(cold || td->td_ucred != NULL,
 542                     ("kernel trap doesn't have ucred"));
 543                 switch (type) {
 544                 case T_PAGEFLT:                 /* page fault */
 545                         (void) trap_pfault(frame, FALSE, eva);
 546                         goto out;
 547
 548                 case T_DNA:
 549 #ifdef DEV_NPX
 550                         if (PCB_USER_FPU(td->td_pcb))
 551                                 panic("Unregistered use of FPU in kernel");
 552                         if (npxdna())
 553                                 goto out;
 554 #endif
 555                         break;
 556
 557                 case T_ARITHTRAP:       /* arithmetic trap */
 558                 case T_XMMFLT:          /* SIMD floating-point exception */
 559                 case T_FPOPFLT:         /* FPU operand fetch fault */
 560                         /*
 561                          * XXXKIB for now disable any FPU traps in kernel
 562                          * handler registration seems to be overkill
 563                          */
 564                         trap_fatal(frame, 0);
 565                         goto out;
 566
 567                         /*
 568                          * The following two traps can happen in
 569                          * vm86 mode, and, if so, we want to handle
 570                          * them specially.
 571                          */
 572                 case T_PROTFLT:         /* general protection fault */
 573                 case T_STKFLT:          /* stack fault */
 574                         if (frame->tf_eflags & PSL_VM) {
 575                                 i = vm86_emulate((struct vm86frame *)frame);
 576                                 if (i == SIGTRAP) {
 577                                         type = T_TRCTRAP;
 578                                         load_dr6(rdr6() | 0x4000);
 579                                         goto kernel_trctrap;
 580                                 }
 581                                 if (i != 0)
 582                                         /*
 583                                          * returns to original process
 584                                          */
 585                                         vm86_trap((struct vm86frame *)frame);
 586                                 goto out;
 587                         }
 588                         if (type == T_STKFLT)
 589                                 break;
 590
 591                         /* FALL THROUGH */
 592
 593                 case T_SEGNPFLT:        /* segment not present fault */
 594                         if (curpcb->pcb_flags & PCB_VM86CALL)
 595                                 break;
 596
 597                         /*
 598                          * Invalid %fs's and %gs's can be created using
 599                          * procfs or PT_SETREGS or by invalidating the
 600                          * underlying LDT entry.  This causes a fault
 601                          * in kernel mode when the kernel attempts to
 602                          * switch contexts.  Lose the bad context
 603                          * (XXX) so that we can continue, and generate
 604                          * a signal.
 605                          */
 606                         if (frame->tf_eip == (int)cpu_switch_load_gs) {
 607                                 curpcb->pcb_gs = 0;
 608 #if 0
 609                                 PROC_LOCK(p);
 610                                 kern_psignal(p, SIGBUS);
 611                                 PROC_UNLOCK(p);
 612 #endif
 613                                 goto out;
 614                         }
 615
 616                         if (td->td_intr_nesting_level != 0)
 617                                 break;
 618
 619                         /*
 620                          * Invalid segment selectors and out of bounds
 621                          * %eip's and %esp's can be set up in user mode.
 622                          * This causes a fault in kernel mode when the
 623                          * kernel tries to return to user mode.  We want
 624                          * to get this fault so that we can fix the
 625                          * problem here and not have to check all the
 626                          * selectors and pointers when the user changes
 627                          * them.
 628                          */
 629                         if (frame->tf_eip == (int)doreti_iret) {
 630                                 frame->tf_eip = (int)doreti_iret_fault;
 631                                 goto out;
 632                         }
 633                         if (frame->tf_eip == (int)doreti_popl_ds) {
 634                                 frame->tf_eip = (int)doreti_popl_ds_fault;
 635                                 goto out;
 636                         }
 637                         if (frame->tf_eip == (int)doreti_popl_es) {
 638                                 frame->tf_eip = (int)doreti_popl_es_fault;
 639                                 goto out;
 640                         }
 641                         if (frame->tf_eip == (int)doreti_popl_fs) {
 642                                 frame->tf_eip = (int)doreti_popl_fs_fault;
 643                                 goto out;
 644                         }
 645                         if (curpcb->pcb_onfault != NULL) {
 646                                 frame->tf_eip =
 647                                     (int)curpcb->pcb_onfault;
 648                                 goto out;
 649                         }
 650                         break;
 651
 652                 case T_TSSFLT:
 653                         /*
 654                          * PSL_NT can be set in user mode and isn't cleared
 655                          * automatically when the kernel is entered.  This
 656                          * causes a TSS fault when the kernel attempts to
 657                          * `iret' because the TSS link is uninitialized.  We
 658                          * want to get this fault so that we can fix the
 659                          * problem here and not every time the kernel is
 660                          * entered.
 661                          */
 662                         if (frame->tf_eflags & PSL_NT) {
 663                                 frame->tf_eflags &= ~PSL_NT;
 664                                 goto out;
 665                         }
 666                         break;
 667
 668                 case T_TRCTRAP:  /* trace trap */
 669 kernel_trctrap:
 670                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall)) {
 671                                 /*
 672                                  * We've just entered system mode via the
 673                                  * syscall lcall.  Continue single stepping
 674                                  * silently until the syscall handler has
 675                                  * saved the flags.
 676                                  */
 677                                 goto out;
 678                         }
 679                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall) + 1) {
 680                                 /*
 681                                  * The syscall handler has now saved the
 682                                  * flags.  Stop single stepping it.
 683                                  */
 684                                 frame->tf_eflags &= ~PSL_T;
 685                                 goto out;
 686                         }
 687                         /*
 688                          * Ignore debug register trace traps due to
 689                          * accesses in the user's address space, which
 690                          * can happen under several conditions such as
 691                          * if a user sets a watchpoint on a buffer and
 692                          * then passes that buffer to a system call.
 693                          * We still want to get TRCTRAPS for addresses
 694                          * in kernel space because that is useful when
 695                          * debugging the kernel.
 696                          */
 697                         if (user_dbreg_trap() &&
 698                            !(curpcb->pcb_flags & PCB_VM86CALL)) {
 699                                 /*
 700                                  * Reset breakpoint bits because the
 701                                  * processor doesn't
 702                                  */
 703                                 load_dr6(rdr6() & ~0xf);
 704                                 goto out;
 705                         }
 706                         /*
 707                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 708                          */
 709                 case T_BPTFLT:
 710                         /*
 711                          * If KDB is enabled, let it handle the debugger trap.
 712                          * Otherwise, debugger traps "can't happen".
 713                          */
 714 #ifdef KDB
 715                         /* XXX %dr6 is not quite reentrant. */
 716                         dr6 = rdr6();
 717                         load_dr6(dr6 & ~0x4000);
 718                         if (kdb_trap(type, dr6, frame))
 719                                 goto out;
 720 #endif
 721                         break;
 722
 723 #ifdef DEV_ISA
 724                 case T_NMI:
 725 #ifdef POWERFAIL_NMI
 726                         if (time_second - lastalert > 10) {
 727                                 log(LOG_WARNING, "NMI: power fail\n");
 728                                 sysbeep(880, hz);
 729                                 lastalert = time_second;
 730                         }
 731                         goto out;
 732 #else /* !POWERFAIL_NMI */
 733                         /* machine/parity/power fail/"kitchen sink" faults */
 734                         if (isa_nmi(frame->tf_err) == 0) {
 735 #ifdef KDB
 736                                 /*
 737                                  * NMI can be hooked up to a pushbutton
 738                                  * for debugging.
 739                                  */
 740                                 if (kdb_on_nmi) {
 741                                         printf ("NMI ... going to debugger\n");
 742                                         kdb_trap(type, 0, frame);
 743                                 }
 744 #endif /* KDB */
 745                                 goto out;
 746                         } else if (panic_on_nmi == 0)
 747                                 goto out;
 748                         /* FALLTHROUGH */
 749 #endif /* POWERFAIL_NMI */
 750 #endif /* DEV_ISA */
 751                 }
 752
 753                 trap_fatal(frame, eva);
 754                 goto out;
 755         }
 756
 757         /* Translate fault for emulators (e.g. Linux) */
 758         if (*p->p_sysent->sv_transtrap)
 759                 i = (*p->p_sysent->sv_transtrap)(i, type);
 760
 761         ksiginfo_init_trap(&ksi);
 762         ksi.ksi_signo = i;
 763         ksi.ksi_code = ucode;
 764         ksi.ksi_addr = (void *)addr;
 765         ksi.ksi_trapno = type;
 766         if (uprintf_signal) {
 767                 uprintf("pid %d comm %s: signal %d err %x code %d type %d "
 768                     "addr 0x%x esp 0x%08x eip 0x%08x "
 769                     "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
 770                     p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr,
 771                     frame->tf_esp, frame->tf_eip,
 772                     fubyte((void *)(frame->tf_eip + 0)),
 773                     fubyte((void *)(frame->tf_eip + 1)),
 774                     fubyte((void *)(frame->tf_eip + 2)),
 775                     fubyte((void *)(frame->tf_eip + 3)),
 776                     fubyte((void *)(frame->tf_eip + 4)),
 777                     fubyte((void *)(frame->tf_eip + 5)),
 778                     fubyte((void *)(frame->tf_eip + 6)),
 779                     fubyte((void *)(frame->tf_eip + 7)));
 780         }
 781         KASSERT((read_eflags() & PSL_I) != 0, ("interrupts disabled"));
 782         trapsignal(td, &ksi);
 783
 784 #ifdef DEBUG
 785         if (type <= MAX_TRAP_MSG) {
 786                 uprintf("fatal process exception: %s",
 787                         trap_msg[type]);
 788                 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
 789                         uprintf(", fault VA = 0x%lx", (u_long)eva);
 790                 uprintf("\n");
 791         }
 792 #endif
 793
 794 user:
 795         userret(td, frame);
 796         KASSERT(PCB_USER_FPU(td->td_pcb),
 797             ("Return from trap with kernel FPU ctx leaked"));
 798 userout:
 799 out:
 800         return;
 801 }
 802
 803 static int
 804 trap_pfault(frame, usermode, eva)
 805         struct trapframe *frame;
 806         int usermode;
 807         vm_offset_t eva;
 808 {
 809         vm_offset_t va;
 810         vm_map_t map;
 811         int rv = 0;
 812         vm_prot_t ftype;
 813         struct thread *td = curthread;
 814         struct proc *p = td->td_proc;
 815
 816         if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
 817                 /*
 818                  * Due to both processor errata and lazy TLB invalidation when
 819                  * access restrictions are removed from virtual pages, memory
 820                  * accesses that are allowed by the physical mapping layer may
 821                  * nonetheless cause one spurious page fault per virtual page.
 822                  * When the thread is executing a "no faulting" section that
 823                  * is bracketed by vm_fault_{disable,enable}_pagefaults(),
 824                  * every page fault is treated as a spurious page fault,
 825                  * unless it accesses the same virtual address as the most
 826                  * recent page fault within the same "no faulting" section.
 827                  */
 828                 if (td->td_md.md_spurflt_addr != eva ||
 829                     (td->td_pflags & TDP_RESETSPUR) != 0) {
 830                         /*
 831                          * Do nothing to the TLB.  A stale TLB entry is
 832                          * flushed automatically by a page fault.
 833                          */
 834                         td->td_md.md_spurflt_addr = eva;
 835                         td->td_pflags &= ~TDP_RESETSPUR;
 836                         return (0);
 837                 }
 838         } else {
 839                 /*
 840                  * If we get a page fault while in a critical section, then
 841                  * it is most likely a fatal kernel page fault.  The kernel
 842                  * is already going to panic trying to get a sleep lock to
 843                  * do the VM lookup, so just consider it a fatal trap so the
 844                  * kernel can print out a useful trap message and even get
 845                  * to the debugger.
 846                  *
 847                  * If we get a page fault while holding a non-sleepable
 848                  * lock, then it is most likely a fatal kernel page fault.
 849                  * If WITNESS is enabled, then it's going to whine about
 850                  * bogus LORs with various VM locks, so just skip to the
 851                  * fatal trap handling directly.
 852                  */
 853                 if (td->td_critnest != 0 ||
 854                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 855                     "Kernel page fault") != 0) {
 856                         trap_fatal(frame, eva);
 857                         return (-1);
 858                 }
 859         }
 860         va = trunc_page(eva);
 861         if (va >= KERNBASE) {
 862                 /*
 863                  * Don't allow user-mode faults in kernel address space.
 864                  * An exception:  if the faulting address is the invalid
 865                  * instruction entry in the IDT, then the Intel Pentium
 866                  * F00F bug workaround was triggered, and we need to
 867                  * treat it is as an illegal instruction, and not a page
 868                  * fault.
 869                  */
 870 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 871                 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
 872                         return (-2);
 873 #endif
 874                 if (usermode)
 875                         goto nogo;
 876
 877                 map = kernel_map;
 878         } else {
 879                 map = &p->p_vmspace->vm_map;
 880
 881                 /*
 882                  * When accessing a user-space address, kernel must be
 883                  * ready to accept the page fault, and provide a
 884                  * handling routine.  Since accessing the address
 885                  * without the handler is a bug, do not try to handle
 886                  * it normally, and panic immediately.
 887                  */
 888                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 889                     curpcb->pcb_onfault == NULL)) {
 890                         trap_fatal(frame, eva);
 891                         return (-1);
 892                 }
 893         }
 894
 895         /*
 896          * PGEX_I is defined only if the execute disable bit capability is
 897          * supported and enabled.
 898          */
 899         if (frame->tf_err & PGEX_W)
 900                 ftype = VM_PROT_WRITE;
 901 #if defined(PAE) || defined(PAE_TABLES)
 902         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 903                 ftype = VM_PROT_EXECUTE;
 904 #endif
 905         else
 906                 ftype = VM_PROT_READ;
 907
 908         /* Fault in the page. */
 909         rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 910         if (rv == KERN_SUCCESS) {
 911 #ifdef HWPMC_HOOKS
 912                 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
 913                         PMC_SOFT_CALL_TF( , , page_fault, all, frame);
 914                         if (ftype == VM_PROT_READ)
 915                                 PMC_SOFT_CALL_TF( , , page_fault, read,
 916                                     frame);
 917                         else
 918                                 PMC_SOFT_CALL_TF( , , page_fault, write,
 919                                     frame);
 920                 }
 921 #endif
 922                 return (0);
 923         }
 924 nogo:
 925         if (!usermode) {
 926                 if (td->td_intr_nesting_level == 0 &&
 927                     curpcb->pcb_onfault != NULL) {
 928                         frame->tf_eip = (int)curpcb->pcb_onfault;
 929                         return (0);
 930                 }
 931                 trap_fatal(frame, eva);
 932                 return (-1);
 933         }
 934         return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 935 }
 936
 937 static void
 938 trap_fatal(frame, eva)
 939         struct trapframe *frame;
 940         vm_offset_t eva;
 941 {
 942         int code, ss, esp;
 943         u_int type;
 944         struct soft_segment_descriptor softseg;
 945         char *msg;
 946
 947         code = frame->tf_err;
 948         type = frame->tf_trapno;
 949         sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
 950
 951         if (type <= MAX_TRAP_MSG)
 952                 msg = trap_msg[type];
 953         else
 954                 msg = "UNKNOWN";
 955         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 956             frame->tf_eflags & PSL_VM ? "vm86" :
 957             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
 958 #ifdef SMP
 959         /* two separate prints in case of a trap on an unmapped page */
 960         printf("cpuid = %d; ", PCPU_GET(cpuid));
 961         printf("apic id = %02x\n", PCPU_GET(apic_id));
 962 #endif
 963         if (type == T_PAGEFLT) {
 964                 printf("fault virtual address   = 0x%x\n", eva);
 965                 printf("fault code              = %s %s, %s\n",
 966                         code & PGEX_U ? "user" : "supervisor",
 967                         code & PGEX_W ? "write" : "read",
 968                         code & PGEX_P ? "protection violation" : "page not present");
 969         }
 970         printf("instruction pointer     = 0x%x:0x%x\n",
 971                frame->tf_cs & 0xffff, frame->tf_eip);
 972         if (TF_HAS_STACKREGS(frame)) {
 973                 ss = frame->tf_ss & 0xffff;
 974                 esp = frame->tf_esp;
 975         } else {
 976                 ss = GSEL(GDATA_SEL, SEL_KPL);
 977                 esp = (int)&frame->tf_esp;
 978         }
 979         printf("stack pointer           = 0x%x:0x%x\n", ss, esp);
 980         printf("frame pointer           = 0x%x:0x%x\n", ss, frame->tf_ebp);
 981         printf("code segment            = base 0x%x, limit 0x%x, type 0x%x\n",
 982                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 983         printf("                        = DPL %d, pres %d, def32 %d, gran %d\n",
 984                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
 985                softseg.ssd_gran);
 986         printf("processor eflags        = ");
 987         if (frame->tf_eflags & PSL_T)
 988                 printf("trace trap, ");
 989         if (frame->tf_eflags & PSL_I)
 990                 printf("interrupt enabled, ");
 991         if (frame->tf_eflags & PSL_NT)
 992                 printf("nested task, ");
 993         if (frame->tf_eflags & PSL_RF)
 994                 printf("resume, ");
 995         if (frame->tf_eflags & PSL_VM)
 996                 printf("vm86, ");
 997         printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
 998         printf("current process         = %d (%s)\n",
 999             curproc->p_pid, curthread->td_name);
1000
1001 #ifdef KDB
1002         if (debugger_on_panic || kdb_active) {
1003                 frame->tf_err = eva;    /* smuggle fault address to ddb */
1004                 if (kdb_trap(type, 0, frame)) {
1005                         frame->tf_err = code;   /* restore error code */
1006                         return;
1007                 }
1008                 frame->tf_err = code;           /* restore error code */
1009         }
1010 #endif
1011         printf("trap number             = %d\n", type);
1012         if (type <= MAX_TRAP_MSG)
1013                 panic("%s", trap_msg[type]);
1014         else
1015                 panic("unknown/reserved trap");
1016 }
1017
1018 /*
1019  * Double fault handler. Called when a fault occurs while writing
1020  * a frame for a trap/exception onto the stack. This usually occurs
1021  * when the stack overflows (such is the case with infinite recursion,
1022  * for example).
1023  *
1024  * XXX Note that the current PTD gets replaced by IdlePTD when the
1025  * task switch occurs. This means that the stack that was active at
1026  * the time of the double fault is not available at <kstack> unless
1027  * the machine was idle when the double fault occurred. The downside
1028  * of this is that "trace <ebp>" in ddb won't work.
1029  */
1030 void
1031 dblfault_handler()
1032 {
1033 #ifdef KDTRACE_HOOKS
1034         if (dtrace_doubletrap_func != NULL)
1035                 (*dtrace_doubletrap_func)();
1036 #endif
1037         printf("\nFatal double fault:\n");
1038         printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip));
1039         printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp));
1040         printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp));
1041 #ifdef SMP
1042         /* two separate prints in case of a trap on an unmapped page */
1043         printf("cpuid = %d; ", PCPU_GET(cpuid));
1044         printf("apic id = %02x\n", PCPU_GET(apic_id));
1045 #endif
1046         panic("double fault");
1047 }
1048
1049 int
1050 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
1051 {
1052         struct proc *p;
1053         struct trapframe *frame;
1054         caddr_t params;
1055         long tmp;
1056         int error;
1057
1058         p = td->td_proc;
1059         frame = td->td_frame;
1060
1061         params = (caddr_t)frame->tf_esp + sizeof(int);
1062         sa->code = frame->tf_eax;
1063
1064         /*
1065          * Need to check if this is a 32 bit or 64 bit syscall.
1066          */
1067         if (sa->code == SYS_syscall) {
1068                 /*
1069                  * Code is first argument, followed by actual args.
1070                  */
1071                 error = fueword(params, &tmp);
1072                 if (error == -1)
1073                         return (EFAULT);
1074                 sa->code = tmp;
1075                 params += sizeof(int);
1076         } else if (sa->code == SYS___syscall) {
1077                 /*
1078                  * Like syscall, but code is a quad, so as to maintain
1079                  * quad alignment for the rest of the arguments.
1080                  */
1081                 error = fueword(params, &tmp);
1082                 if (error == -1)
1083                         return (EFAULT);
1084                 sa->code = tmp;
1085                 params += sizeof(quad_t);
1086         }
1087
1088         if (p->p_sysent->sv_mask)
1089                 sa->code &= p->p_sysent->sv_mask;
1090         if (sa->code >= p->p_sysent->sv_size)
1091                 sa->callp = &p->p_sysent->sv_table[0];
1092         else
1093                 sa->callp = &p->p_sysent->sv_table[sa->code];
1094         sa->narg = sa->callp->sy_narg;
1095
1096         if (params != NULL && sa->narg != 0)
1097                 error = copyin(params, (caddr_t)sa->args,
1098                     (u_int)(sa->narg * sizeof(int)));
1099         else
1100                 error = 0;
1101
1102         if (error == 0) {
1103                 td->td_retval[0] = 0;
1104                 td->td_retval[1] = frame->tf_edx;
1105         }
1106
1107         return (error);
1108 }
1109
1110 #include "../../kern/subr_syscall.c"
1111
1112 /*
1113  * syscall - system call request C handler.  A system call is
1114  * essentially treated as a trap by reusing the frame layout.
1115  */
1116 void
1117 syscall(struct trapframe *frame)
1118 {
1119         struct thread *td;
1120         struct syscall_args sa;
1121         register_t orig_tf_eflags;
1122         int error;
1123         ksiginfo_t ksi;
1124
1125 #ifdef DIAGNOSTIC
1126         if (!(TRAPF_USERMODE(frame) &&
1127             (curpcb->pcb_flags & PCB_VM86CALL) == 0)) {
1128                 panic("syscall");
1129                 /* NOT REACHED */
1130         }
1131 #endif
1132         orig_tf_eflags = frame->tf_eflags;
1133
1134         td = curthread;
1135         td->td_frame = frame;
1136
1137         error = syscallenter(td, &sa);
1138
1139         /*
1140          * Traced syscall.
1141          */
1142         if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1143                 frame->tf_eflags &= ~PSL_T;
1144                 ksiginfo_init_trap(&ksi);
1145                 ksi.ksi_signo = SIGTRAP;
1146                 ksi.ksi_code = TRAP_TRACE;
1147                 ksi.ksi_addr = (void *)frame->tf_eip;
1148                 trapsignal(td, &ksi);
1149         }
1150
1151         KASSERT(PCB_USER_FPU(td->td_pcb),
1152             ("System call %s returning with kernel FPU ctx leaked",
1153              syscallname(td->td_proc, sa.code)));
1154         KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td),
1155             ("System call %s returning with mangled pcb_save",
1156              syscallname(td->td_proc, sa.code)));
1157
1158         syscallret(td, error, &sa);
1159 }