insns.pl

汇编编译器的最新版本的源码.买了自己动手写操作系统这本书的人一定要下

    return $s;
}

sub addprefix ($@) {
    my ($prefix, @list) = @_;
    my $x;
    my @l = ();

    foreach $x (@list) {
	push(@l, sprintf("%s%02X", $prefix, $x));
    }

    return @l;
}

#
# Turn a code string into a sequence of bytes
#
sub decodify($) {
  # Although these are C-syntax strings, by convention they should have
  # only octal escapes (for directives) and hexadecimal escapes
  # (for verbatim bytes)
    my($codestr) = @_;

    if ($codestr =~ /^\s*\[([^\]]*)\]\s*$/) {
	return byte_code_compile($1);
    }

    my $c = $codestr;
    my @codes = ();

    while ($c ne '') {
	if ($c =~ /^\\x([0-9a-f]+)(.*)$/i) {
	    push(@codes, hex $1);
	    $c = $2;
	    next;
	} elsif ($c =~ /^\\([0-7]{1,3})(.*)$/) {
	    push(@codes, oct $1);
	    $c = $2;
	    next;
	} else {
	    die "$fname: unknown code format in \"$codestr\"\n";
	}
    }

    return @codes;
}

# Turn a numeric list into a hex string
sub hexstr(@) {
    my $s = '';
    my $c;

    foreach $c (@_) {
	$s .= sprintf("%02X", $c);
    }
    return $s;
}

# Here we determine the range of possible starting bytes for a given
# instruction. We need only consider the codes:
# \1 \2 \3     mean literal bytes, of course
# \4 \5 \6 \7  mean PUSH/POP of segment registers: special case
# \1[0123]     mean byte plus register value
# \330         means byte plus condition code
# \0 or \340   mean give up and return empty set
# \17[234]     skip is4 control byte
# \26x \270    skip VEX control bytes
sub startseq($) {
  my ($codestr) = @_;
  my $word, @range;
  my @codes = ();
  my $c = $codestr;
  my $c0, $c1, $i;
  my $prefix = '';

  @codes = decodify($codestr);

  while ($c0 = shift(@codes)) {
      $c1 = $codes[0];
      if ($c0 == 01 || $c0 == 02 || $c0 == 03) {
	  # Fixed byte string
	  my $fbs = $prefix;
	  while (1) {
	      if ($c0 == 01 || $c0 == 02 || $c0 == 03) {
		  while ($c0--) {
		      $fbs .= sprintf("%02X", shift(@codes));
		  }
	      } else {
		  last;
	      }
	      $c0 = shift(@codes);
	  }

	  foreach $pfx (@disasm_prefixes) {
	      if (substr($fbs, 0, length($pfx)) eq $pfx) {
		  $prefix = $pfx;
		  $fbs = substr($fbs, length($pfx));
		  last;
	      }
	  }

	  if ($fbs ne '') {
	      return ($prefix.substr($fbs,0,2));
	  }

	  unshift(@codes, $c0);
      } elsif ($c0 >= 010 && $c0 <= 013) {
	  return addprefix($prefix, $c1..($c1+7));
      } elsif (($c0 & ~013) == 0144) {
	  return addprefix($prefix, $c1, $c1|2);
      } elsif ($c0 == 0330) {
	  return addprefix($prefix, $c1..($c1+15));
      } elsif ($c0 == 0 || $c0 == 0340) {
	  return $prefix;
      } elsif ($c0 == 0344) {
	  return addprefix($prefix, 0x06, 0x0E, 0x16, 0x1E);
      } elsif ($c0 == 0345) {
	  return addprefix($prefix, 0x07, 0x17, 0x1F);
      } elsif ($c0 == 0346) {
	  return addprefix($prefix, 0xA0, 0xA8);
      } elsif ($c0 == 0347) {
	  return addprefix($prefix, 0xA1, 0xA9);
      } elsif (($c0 & ~3) == 0260 || $c0 == 0270) {
	  my $m,$wlp,$vxp;
	  $m   = shift(@codes);
	  $wlp = shift(@codes);
	  $prefix .= sprintf('VEX%02X%01X', $m, $wlp & 7);
      } elsif ($c0 >= 0172 && $c0 <= 174) {
	  shift(@codes);	# Skip is4 control byte
      } else {
	  # We really need to be able to distinguish "forbidden"
	  # and "ignorable" codes here
      }
  }
  return $prefix;
}

#
# This function takes a series of byte codes in a format which is more
# typical of the Intel documentation, and encode it.
#
# The format looks like:
#
# [operands: opcodes]
#
# The operands word lists the order of the operands:
#
# r = register field in the modr/m
# m = modr/m
# v = VEX "v" field
# d = DREX "dst" field
# i = immediate
# s = register field of is4/imz2 field
# - = implicit (unencoded) operand
#
# For an operand that should be filled into more than one field,
# enter it as e.g. "r+v".
#
sub byte_code_compile($) {
    my($str) = @_;
    my $opr;
    my $opc;
    my @codes = ();
    my $litix = undef;
    my %oppos = ();
    my $i;
    my $op, $oq;

    unless ($str =~ /^(([^\s:]*)\:|)\s*(.*\S)\s*$/) {
	die "$fname: $line: cannot parse: [$str]\n";
    }
    $opr = "\L$2";
    $opc = "\L$3";

    my $op = 0;
    for ($i = 0; $i < length($opr); $i++) {
	my $c = substr($opr,$i,1);
	if ($c eq '+') {
	    $op--;
	} else {
	    $oppos{$c} = $op++;
	}
    }

    $prefix_ok = 1;
    foreach $op (split(/\s*(?:\s|(?=[\/\\]))/, $opc)) {
	if ($op eq 'o16') {
	    push(@codes, 0320);
	} elsif ($op eq 'o32') {
	    push(@codes, 0321);
	} elsif ($op eq 'o64') {  # 64-bit operand size requiring REX.W
	    push(@codes, 0324);
	} elsif ($op eq 'o64nw') { # Implied 64-bit operand size (no REX.W)
	    push(@codes, 0323);
	} elsif ($op eq 'a16') {
	    push(@codes, 0310);
	} elsif ($op eq 'a32') {
	    push(@codes, 0311);
	} elsif ($op eq 'a64') {
	    push(@codes, 0313);
	} elsif ($op eq '!osp') {
	    push(@codes, 0364);
	} elsif ($op eq '!asp') {
	    push(@codes, 0365);
	} elsif ($op eq 'rex.l') {
	    push(@codes, 0334);
	} elsif ($op eq 'repe') {
	    push(@codes, 0335);
	} elsif ($prefix_ok && $op =~ /^(66|f2|f3|np)$/) {
	    # 66/F2/F3 prefix used as an opcode extension, or np = no prefix
	    if ($op eq '66') {
		push(@codes, 0361);
	    } elsif ($op eq 'f2') {
		push(@codes, 0362);
	    } elsif ($op eq 'f3') {
		push(@codes, 0363);
	    } else {
		push(@codes, 0360);
	    }
	} elsif ($op =~ /^[0-9a-f]{2}$/) {
	    if (defined($litix) && $litix+$codes[$litix]+1 == scalar @codes) {
		$codes[$litix]++;
		push(@codes, hex $op);
	    } else {
		$litix = scalar(@codes);
		push(@codes, 01, hex $op);
	    }
	    $prefix_ok = 0;
	} elsif ($op eq '/r') {
	    if (!defined($oppos{'r'}) || !defined($oppos{'m'})) {
		die "$fname: $line: $op requires r and m operands\n";
	    }
	    push(@codes, 0100 + ($oppos{'m'} << 3) + $oppos{'r'});
	    $prefix_ok = 0;
	} elsif ($op =~ m:^/([0-7])$:) {
	    if (!defined($oppos{'m'})) {
		die "$fname: $line: $op requires m operand\n";
	    }
	    push(@codes, 0200 + ($oppos{'m'} << 3) + $1);
	    $prefix_ok = 0;
	} elsif ($op =~ /^vex(|\..*)$/) {
	    my ($m,$w,$l,$p) = (undef,2,undef,0);
	    my $has_nds = 0;
	    foreach $oq (split(/\./, $op)) {
		if ($oq eq 'vex') {
		    # prefix
		} elsif ($oq eq '128' || $oq eq 'l0') {
		    $l = 0;
		} elsif ($oq eq '256' || $oq eq 'l1') {
		    $l = 1;
		} elsif ($oq eq 'w0') {
		    $w = 0;
		} elsif ($oq eq 'w1') {
		    $w = 1;
		} elsif ($oq eq 'wx') {
		    $w = 2;
		} elsif ($oq eq 'ww') {
		    $w = 3;
		} elsif ($oq eq '66') {
		    $p = 1;
		} elsif ($oq eq 'f3') {
		    $p = 2;
		} elsif ($oq eq 'f2') {
		    $p = 3;
		} elsif ($oq eq '0f') {
		    $m = 1;
		} elsif ($oq eq '0f38') {
		    $m = 2;
		} elsif ($oq eq '0f3a') {
		    $m = 3;
		} elsif ($oq =~ /^m([0-9]+)$/) {
		    $m = $1+0;
		} elsif ($oq eq 'nds' || $oq eq 'ndd') {
		    if (!defined($oppos{'v'})) {
			die "$fname: $line: vex.$oq without 'v' operand\n";
		    }
		    $has_nds = 1;
		} else {
		    die "$fname: $line: undefined VEX subcode: $oq\n";
		}
	    }
	    if (!defined($m) || !defined($w) || !defined($l) || !defined($p)) {
		die "$fname: $line: missing fields in VEX specification\n";
	    }
	    if (defined($oppos{'v'}) && !$has_nds) {
		die "$fname: $line: 'v' operand without vex.nds or vex.ndd\n";
	    }
	    push(@codes, defined($oppos{'v'}) ? 0260+$oppos{'v'} : 0270,
		 $m, ($w << 3)+($l << 2)+$p);
	    $prefix_ok = 0;
	} elsif ($op =~ /^\/drex([01])$/) {
	    my $oc0 = $1;
	    if (!defined($oppos{'d'})) {
		die "$fname: $line: DREX without a 'd' operand\n";
	    }
	    # Note the use of *unshift* here, as opposed to *push*.
	    # This is because NASM want this byte code at the start of
	    # the instruction sequence, but the AMD documentation puts
	    # this at (roughly) the position of the drex byte itself.
	    # This allows us to match the AMD documentation and still
	    # do the right thing.
	    unshift(@codes, 0160+$oppos{'d'}+($oc0 ? 4 : 0));
	} elsif ($op =~ /^(ib\,s|ib|ibx|ib\,w|iw|iwd|id|idx|iwdq|rel|rel8|rel16|rel32|iq|seg|ibw|ibd|ibd,s)$/) {
	    if (!defined($oppos{'i'})) {
		die "$fname: $line: $op without 'i' operand\n";
	    }
	    if ($op eq 'ib,s') { # Signed imm8
		push(@codes, 014+$oppos{'i'});
	    } elsif ($op eq 'ib') { # imm8
		push(@codes, 020+$oppos{'i'});
	    } elsif ($op eq 'ib,u') { # Unsigned imm8
		push(@codes, 024+$oppos{'i'});
	    } elsif ($op eq 'iw') { # imm16
		push(@codes, 030+$oppos{'i'});
	    } elsif ($op eq 'ibx') { # imm8 sign-extended to opsize
		push(@codes, 0274+$oppos{'i'});
	    } elsif ($op eq 'iwd') { # imm16 or imm32, depending on opsize
		push(@codes, 034+$oppos{'i'});
	    } elsif ($op eq 'id') { # imm32
		push(@codes, 040+$oppos{'i'});
	    } elsif ($op eq 'idx') { # imm32 extended to 64 bits
		push(@codes, 0254+$oppos{'i'});
	    } elsif ($op eq 'iwdq') { # imm16/32/64, depending on opsize
		push(@codes, 044+$oppos{'i'});
	    } elsif ($op eq 'rel8') {
		push(@codes, 050+$oppos{'i'});
	    } elsif ($op eq 'iq') {
		push(@codes, 054+$oppos{'i'});
	    } elsif ($op eq 'rel16') {
		push(@codes, 060+$oppos{'i'});
	    } elsif ($op eq 'rel') { # 16 or 32 bit relative operand
		push(@codes, 064+$oppos{'i'});
	    } elsif ($op eq 'rel32') {
		push(@codes, 070+$oppos{'i'});
	    } elsif ($op eq 'seg') {
		push(@codes, 074+$oppos{'i'});
	    } elsif ($op eq 'ibw') { # imm16 that can be bytified
		if (!defined($s_pos)) {
		    die "$fname: $line: $op without a +s byte\n";
		}
		$codes[$s_pos] += 0144;
		push(@codes, 0140+$oppos{'i'});
	    } elsif ($op eq 'ibd') { # imm32 that can be bytified
		if (!defined($s_pos)) {
		    die "$fname: $line: $op without a +s byte\n";
		}
		$codes[$s_pos] += 0154;
		push(@codes, 0150+$oppos{'i'});
	    } elsif ($op eq 'ibd,s') {
		# imm32 that can be bytified, sign extended to 64 bits
		if (!defined($s_pos)) {
		    die "$fname: $line: $op without a +s byte\n";
		}
		$codes[$s_pos] += 0154;
		push(@codes, 0250+$oppos{'i'});
	    }
	    $prefix_ok = 0;
	} elsif ($op eq '/is4') {
	    if (!defined($oppos{'s'})) {
		die "$fname: $line: $op without 's' operand\n";
	    }
	    if (defined($oppos{'i'})) {
		push(@codes, 0172, ($oppos{'s'} << 3)+$oppos{'i'});
	    } else {
		push(@codes, 0174, $oppos{'s'});
	    }
	    $prefix_ok = 0;
	} elsif ($op =~ /^\/is4\=([0-9]+)$/) {
	    my $imm = $1;
	    if (!defined($oppos{'s'})) {
		die "$fname: $line: $op without 's' operand\n";
	    }
	    if ($imm < 0 || $imm > 15) {
		die "$fname: $line: invalid imm4 value for $op: $imm\n";
	    }
	    push(@codes, 0173, ($oppos{'s'} << 4) + $imm);
	    $prefix_ok = 0;
	} elsif ($op =~ /^([0-9a-f]{2})\+s$/) {
	    if (!defined($oppos{'i'})) {
		die "$fname: $line: $op without 'i' operand\n";
	    }
	    $s_pos = scalar @codes;
	    push(@codes, $oppos{'i'}, hex $1);
	    $prefix_ok = 0;
	} elsif ($op =~ /^([0-9a-f]{2})\+c$/) {
	    push(@codes, 0330, hex $1);
	    $prefix_ok = 0;
	} elsif ($op =~ /^\\([0-7]+|x[0-9a-f]{2})$/) {
	    # Escape to enter literal bytecodes
	    push(@codes, oct $1);
	} else {
	    die "$fname: $line: unknown operation: $op\n";
	}
    }

    return @codes;
}