| 1 | #! /usr/bin/env perl
|
|---|
| 2 | # Copyright 2005-2020 The OpenSSL Project Authors. All Rights Reserved.
|
|---|
| 3 | #
|
|---|
| 4 | # Licensed under the OpenSSL license (the "License"). You may not use
|
|---|
| 5 | # this file except in compliance with the License. You can obtain a copy
|
|---|
| 6 | # in the file LICENSE in the source distribution or at
|
|---|
| 7 | # https://www.openssl.org/source/license.html
|
|---|
| 8 |
|
|---|
| 9 |
|
|---|
| 10 | # Ascetic x86_64 AT&T to MASM/NASM assembler translator by <appro>.
|
|---|
| 11 | #
|
|---|
| 12 | # Why AT&T to MASM and not vice versa? Several reasons. Because AT&T
|
|---|
| 13 | # format is way easier to parse. Because it's simpler to "gear" from
|
|---|
| 14 | # Unix ABI to Windows one [see cross-reference "card" at the end of
|
|---|
| 15 | # file]. Because Linux targets were available first...
|
|---|
| 16 | #
|
|---|
| 17 | # In addition the script also "distills" code suitable for GNU
|
|---|
| 18 | # assembler, so that it can be compiled with more rigid assemblers,
|
|---|
| 19 | # such as Solaris /usr/ccs/bin/as.
|
|---|
| 20 | #
|
|---|
| 21 | # This translator is not designed to convert *arbitrary* assembler
|
|---|
| 22 | # code from AT&T format to MASM one. It's designed to convert just
|
|---|
| 23 | # enough to provide for dual-ABI OpenSSL modules development...
|
|---|
| 24 | # There *are* limitations and you might have to modify your assembler
|
|---|
| 25 | # code or this script to achieve the desired result...
|
|---|
| 26 | #
|
|---|
| 27 | # Currently recognized limitations:
|
|---|
| 28 | #
|
|---|
| 29 | # - can't use multiple ops per line;
|
|---|
| 30 | #
|
|---|
| 31 | # Dual-ABI styling rules.
|
|---|
| 32 | #
|
|---|
| 33 | # 1. Adhere to Unix register and stack layout [see cross-reference
|
|---|
| 34 | # ABI "card" at the end for explanation].
|
|---|
| 35 | # 2. Forget about "red zone," stick to more traditional blended
|
|---|
| 36 | # stack frame allocation. If volatile storage is actually required
|
|---|
| 37 | # that is. If not, just leave the stack as is.
|
|---|
| 38 | # 3. Functions tagged with ".type name,@function" get crafted with
|
|---|
| 39 | # unified Win64 prologue and epilogue automatically. If you want
|
|---|
| 40 | # to take care of ABI differences yourself, tag functions as
|
|---|
| 41 | # ".type name,@abi-omnipotent" instead.
|
|---|
| 42 | # 4. To optimize the Win64 prologue you can specify number of input
|
|---|
| 43 | # arguments as ".type name,@function,N." Keep in mind that if N is
|
|---|
| 44 | # larger than 6, then you *have to* write "abi-omnipotent" code,
|
|---|
| 45 | # because >6 cases can't be addressed with unified prologue.
|
|---|
| 46 | # 5. Name local labels as .L*, do *not* use dynamic labels such as 1:
|
|---|
| 47 | # (sorry about latter).
|
|---|
| 48 | # 6. Don't use [or hand-code with .byte] "rep ret." "ret" mnemonic is
|
|---|
| 49 | # required to identify the spots, where to inject Win64 epilogue!
|
|---|
| 50 | # But on the pros, it's then prefixed with rep automatically:-)
|
|---|
| 51 | # 7. Stick to explicit ip-relative addressing. If you have to use
|
|---|
| 52 | # GOTPCREL addressing, stick to mov symbol@GOTPCREL(%rip),%r??.
|
|---|
| 53 | # Both are recognized and translated to proper Win64 addressing
|
|---|
| 54 | # modes.
|
|---|
| 55 | #
|
|---|
| 56 | # 8. In order to provide for structured exception handling unified
|
|---|
| 57 | # Win64 prologue copies %rsp value to %rax. For further details
|
|---|
| 58 | # see SEH paragraph at the end.
|
|---|
| 59 | # 9. .init segment is allowed to contain calls to functions only.
|
|---|
| 60 | # a. If function accepts more than 4 arguments *and* >4th argument
|
|---|
| 61 | # is declared as non 64-bit value, do clear its upper part.
|
|---|
| 62 | |
|---|
| 63 |
|
|---|
| 64 |
|
|---|
| 65 | use strict;
|
|---|
| 66 |
|
|---|
| 67 | my $flavour = shift;
|
|---|
| 68 | my $output = shift;
|
|---|
| 69 | if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
|
|---|
| 70 |
|
|---|
| 71 | open STDOUT,">$output" || die "can't open $output: $!"
|
|---|
| 72 | if (defined($output));
|
|---|
| 73 |
|
|---|
| 74 | my $gas=1; $gas=0 if ($output =~ /\.asm$/);
|
|---|
| 75 | my $elf=1; $elf=0 if (!$gas);
|
|---|
| 76 | my $win64=0;
|
|---|
| 77 | my $prefix="";
|
|---|
| 78 | my $decor=".L";
|
|---|
| 79 |
|
|---|
| 80 | my $masmref=8 + 50727*2**-32; # 8.00.50727 shipped with VS2005
|
|---|
| 81 | my $masm=0;
|
|---|
| 82 | my $PTR=" PTR";
|
|---|
| 83 |
|
|---|
| 84 | my $nasmref=2.03;
|
|---|
| 85 | my $nasm=0;
|
|---|
| 86 |
|
|---|
| 87 | if ($flavour eq "mingw64") { $gas=1; $elf=0; $win64=1;
|
|---|
| 88 | $prefix=`echo __USER_LABEL_PREFIX__ | $ENV{CC} -E -P -`;
|
|---|
| 89 | $prefix =~ s|\R$||; # Better chomp
|
|---|
| 90 | }
|
|---|
| 91 | elsif ($flavour eq "macosx") { $gas=1; $elf=0; $prefix="_"; $decor="L\$"; }
|
|---|
| 92 | elsif ($flavour eq "masm") { $gas=0; $elf=0; $masm=$masmref; $win64=1; $decor="\$L\$"; }
|
|---|
| 93 | elsif ($flavour eq "nasm") { $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $decor="\$L\$"; $PTR=""; }
|
|---|
| 94 | elsif (!$gas)
|
|---|
| 95 | { if ($ENV{ASM} =~ m/nasm/ && `nasm -v` =~ m/version ([0-9]+)\.([0-9]+)/i)
|
|---|
| 96 | { $nasm = $1 + $2*0.01; $PTR=""; }
|
|---|
| 97 | elsif (`ml64 2>&1` =~ m/Version ([0-9]+)\.([0-9]+)(\.([0-9]+))?/)
|
|---|
| 98 | { $masm = $1 + $2*2**-16 + $4*2**-32; }
|
|---|
| 99 | die "no assembler found on %PATH%" if (!($nasm || $masm));
|
|---|
| 100 | $win64=1;
|
|---|
| 101 | $elf=0;
|
|---|
| 102 | $decor="\$L\$";
|
|---|
| 103 | }
|
|---|
| 104 |
|
|---|
| 105 | my $current_segment;
|
|---|
| 106 | my $current_function;
|
|---|
| 107 | my %globals;
|
|---|
| 108 |
|
|---|
| 109 | { package opcode; # pick up opcodes
|
|---|
| 110 | sub re {
|
|---|
| 111 | my ($class, $line) = @_;
|
|---|
| 112 | my $self = {};
|
|---|
| 113 | my $ret;
|
|---|
| 114 |
|
|---|
| 115 | if ($$line =~ /^([a-z][a-z0-9]*)/i) {
|
|---|
| 116 | bless $self,$class;
|
|---|
| 117 | $self->{op} = $1;
|
|---|
| 118 | $ret = $self;
|
|---|
| 119 | $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
|
|---|
| 120 |
|
|---|
| 121 | undef $self->{sz};
|
|---|
| 122 | if ($self->{op} =~ /^(movz)x?([bw]).*/) { # movz is pain...
|
|---|
| 123 | $self->{op} = $1;
|
|---|
| 124 | $self->{sz} = $2;
|
|---|
| 125 | } elsif ($self->{op} =~ /call|jmp/) {
|
|---|
| 126 | $self->{sz} = "";
|
|---|
| 127 | } elsif ($self->{op} =~ /^p/ && $' !~ /^(ush|op|insrw)/) { # SSEn
|
|---|
| 128 | $self->{sz} = "";
|
|---|
| 129 | } elsif ($self->{op} =~ /^[vk]/) { # VEX or k* such as kmov
|
|---|
| 130 | $self->{sz} = "";
|
|---|
| 131 | } elsif ($self->{op} =~ /mov[dq]/ && $$line =~ /%xmm/) {
|
|---|
| 132 | $self->{sz} = "";
|
|---|
| 133 | } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) {
|
|---|
| 134 | $self->{op} = $1;
|
|---|
| 135 | $self->{sz} = $2;
|
|---|
| 136 | }
|
|---|
| 137 | }
|
|---|
| 138 | $ret;
|
|---|
| 139 | }
|
|---|
| 140 | sub size {
|
|---|
| 141 | my ($self, $sz) = @_;
|
|---|
| 142 | $self->{sz} = $sz if (defined($sz) && !defined($self->{sz}));
|
|---|
| 143 | $self->{sz};
|
|---|
| 144 | }
|
|---|
| 145 | sub out {
|
|---|
| 146 | my $self = shift;
|
|---|
| 147 | if ($gas) {
|
|---|
| 148 | if ($self->{op} eq "movz") { # movz is pain...
|
|---|
| 149 | sprintf "%s%s%s",$self->{op},$self->{sz},shift;
|
|---|
| 150 | } elsif ($self->{op} =~ /^set/) {
|
|---|
| 151 | "$self->{op}";
|
|---|
| 152 | } elsif ($self->{op} eq "ret") {
|
|---|
| 153 | my $epilogue = "";
|
|---|
| 154 | if ($win64 && $current_function->{abi} eq "svr4") {
|
|---|
| 155 | $epilogue = "movq 8(%rsp),%rdi\n\t" .
|
|---|
| 156 | "movq 16(%rsp),%rsi\n\t";
|
|---|
| 157 | }
|
|---|
| 158 | $epilogue . ".byte 0xf3,0xc3";
|
|---|
| 159 | } elsif ($self->{op} eq "call" && !$elf && $current_segment eq ".init") {
|
|---|
| 160 | ".p2align\t3\n\t.quad";
|
|---|
| 161 | } else {
|
|---|
| 162 | "$self->{op}$self->{sz}";
|
|---|
| 163 | }
|
|---|
| 164 | } else {
|
|---|
| 165 | $self->{op} =~ s/^movz/movzx/;
|
|---|
| 166 | if ($self->{op} eq "ret") {
|
|---|
| 167 | $self->{op} = "";
|
|---|
| 168 | if ($win64 && $current_function->{abi} eq "svr4") {
|
|---|
| 169 | $self->{op} = "mov rdi,QWORD$PTR\[8+rsp\]\t;WIN64 epilogue\n\t".
|
|---|
| 170 | "mov rsi,QWORD$PTR\[16+rsp\]\n\t";
|
|---|
| 171 | }
|
|---|
| 172 | $self->{op} .= "DB\t0F3h,0C3h\t\t;repret";
|
|---|
| 173 | } elsif ($self->{op} =~ /^(pop|push)f/) {
|
|---|
| 174 | $self->{op} .= $self->{sz};
|
|---|
| 175 | } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") {
|
|---|
| 176 | $self->{op} = "\tDQ";
|
|---|
| 177 | }
|
|---|
| 178 | $self->{op};
|
|---|
| 179 | }
|
|---|
| 180 | }
|
|---|
| 181 | sub mnemonic {
|
|---|
| 182 | my ($self, $op) = @_;
|
|---|
| 183 | $self->{op}=$op if (defined($op));
|
|---|
| 184 | $self->{op};
|
|---|
| 185 | }
|
|---|
| 186 | }
|
|---|
| 187 | { package const; # pick up constants, which start with $
|
|---|
| 188 | sub re {
|
|---|
| 189 | my ($class, $line) = @_;
|
|---|
| 190 | my $self = {};
|
|---|
| 191 | my $ret;
|
|---|
| 192 |
|
|---|
| 193 | if ($$line =~ /^\$([^,]+)/) {
|
|---|
| 194 | bless $self, $class;
|
|---|
| 195 | $self->{value} = $1;
|
|---|
| 196 | $ret = $self;
|
|---|
| 197 | $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
|
|---|
| 198 | }
|
|---|
| 199 | $ret;
|
|---|
| 200 | }
|
|---|
| 201 | sub out {
|
|---|
| 202 | my $self = shift;
|
|---|
| 203 |
|
|---|
| 204 | $self->{value} =~ s/\b(0b[0-1]+)/oct($1)/eig;
|
|---|
| 205 | if ($gas) {
|
|---|
| 206 | # Solaris /usr/ccs/bin/as can't handle multiplications
|
|---|
| 207 | # in $self->{value}
|
|---|
| 208 | my $value = $self->{value};
|
|---|
| 209 | no warnings; # oct might complain about overflow, ignore here...
|
|---|
| 210 | $value =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;
|
|---|
| 211 | if ($value =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg) {
|
|---|
| 212 | $self->{value} = $value;
|
|---|
| 213 | }
|
|---|
| 214 | sprintf "\$%s",$self->{value};
|
|---|
| 215 | } else {
|
|---|
| 216 | my $value = $self->{value};
|
|---|
| 217 | $value =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm);
|
|---|
| 218 | sprintf "%s",$value;
|
|---|
| 219 | }
|
|---|
| 220 | }
|
|---|
| 221 | }
|
|---|
| 222 | { package ea; # pick up effective addresses: expr(%reg,%reg,scale)
|
|---|
| 223 |
|
|---|
| 224 | my %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR",
|
|---|
| 225 | l=>"DWORD$PTR", d=>"DWORD$PTR",
|
|---|
| 226 | q=>"QWORD$PTR", o=>"OWORD$PTR",
|
|---|
| 227 | x=>"XMMWORD$PTR", y=>"YMMWORD$PTR",
|
|---|
| 228 | z=>"ZMMWORD$PTR" ) if (!$gas);
|
|---|
| 229 |
|
|---|
| 230 | sub re {
|
|---|
| 231 | my ($class, $line, $opcode) = @_;
|
|---|
| 232 | my $self = {};
|
|---|
| 233 | my $ret;
|
|---|
| 234 |
|
|---|
| 235 | # optional * ----vvv--- appears in indirect jmp/call
|
|---|
| 236 | if ($$line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)((?:{[^}]+})*)/) {
|
|---|
| 237 | bless $self, $class;
|
|---|
| 238 | $self->{asterisk} = $1;
|
|---|
| 239 | $self->{label} = $2;
|
|---|
| 240 | ($self->{base},$self->{index},$self->{scale})=split(/,/,$3);
|
|---|
| 241 | $self->{scale} = 1 if (!defined($self->{scale}));
|
|---|
| 242 | $self->{opmask} = $4;
|
|---|
| 243 | $ret = $self;
|
|---|
| 244 | $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
|
|---|
| 245 |
|
|---|
| 246 | if ($win64 && $self->{label} =~ s/\@GOTPCREL//) {
|
|---|
| 247 | die if ($opcode->mnemonic() ne "mov");
|
|---|
| 248 | $opcode->mnemonic("lea");
|
|---|
| 249 | }
|
|---|
| 250 | $self->{base} =~ s/^%//;
|
|---|
| 251 | $self->{index} =~ s/^%// if (defined($self->{index}));
|
|---|
| 252 | $self->{opcode} = $opcode;
|
|---|
| 253 | }
|
|---|
| 254 | $ret;
|
|---|
| 255 | }
|
|---|
| 256 | sub size {}
|
|---|
| 257 | sub out {
|
|---|
| 258 | my ($self, $sz) = @_;
|
|---|
| 259 |
|
|---|
| 260 | $self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
|
|---|
| 261 | $self->{label} =~ s/\.L/$decor/g;
|
|---|
| 262 |
|
|---|
| 263 | # Silently convert all EAs to 64-bit. This is required for
|
|---|
| 264 | # elder GNU assembler and results in more compact code,
|
|---|
| 265 | # *but* most importantly AES module depends on this feature!
|
|---|
| 266 | $self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
|
|---|
| 267 | $self->{base} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
|
|---|
| 268 |
|
|---|
| 269 | # Solaris /usr/ccs/bin/as can't handle multiplications
|
|---|
| 270 | # in $self->{label}...
|
|---|
| 271 | use integer;
|
|---|
| 272 | $self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;
|
|---|
| 273 | $self->{label} =~ s/\b([0-9]+\s*[\*\/\%]\s*[0-9]+)\b/eval($1)/eg;
|
|---|
| 274 |
|
|---|
| 275 | # Some assemblers insist on signed presentation of 32-bit
|
|---|
| 276 | # offsets, but sign extension is a tricky business in perl...
|
|---|
| 277 | if ((1<<31)<<1) {
|
|---|
| 278 | $self->{label} =~ s/\b([0-9]+)\b/$1<<32>>32/eg;
|
|---|
| 279 | } else {
|
|---|
| 280 | $self->{label} =~ s/\b([0-9]+)\b/$1>>0/eg;
|
|---|
| 281 | }
|
|---|
| 282 |
|
|---|
| 283 | # if base register is %rbp or %r13, see if it's possible to
|
|---|
| 284 | # flip base and index registers [for better performance]
|
|---|
| 285 | if (!$self->{label} && $self->{index} && $self->{scale}==1 &&
|
|---|
| 286 | $self->{base} =~ /(rbp|r13)/) {
|
|---|
| 287 | $self->{base} = $self->{index}; $self->{index} = $1;
|
|---|
| 288 | }
|
|---|
| 289 |
|
|---|
| 290 | if ($gas) {
|
|---|
| 291 | $self->{label} =~ s/^___imp_/__imp__/ if ($flavour eq "mingw64");
|
|---|
| 292 |
|
|---|
| 293 | if (defined($self->{index})) {
|
|---|
| 294 | sprintf "%s%s(%s,%%%s,%d)%s",
|
|---|
| 295 | $self->{asterisk},$self->{label},
|
|---|
| 296 | $self->{base}?"%$self->{base}":"",
|
|---|
| 297 | $self->{index},$self->{scale},
|
|---|
| 298 | $self->{opmask};
|
|---|
| 299 | } else {
|
|---|
| 300 | sprintf "%s%s(%%%s)%s", $self->{asterisk},$self->{label},
|
|---|
| 301 | $self->{base},$self->{opmask};
|
|---|
| 302 | }
|
|---|
| 303 | } else {
|
|---|
| 304 | $self->{label} =~ s/\./\$/g;
|
|---|
| 305 | $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig;
|
|---|
| 306 | $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/);
|
|---|
| 307 |
|
|---|
| 308 | my $mnemonic = $self->{opcode}->mnemonic();
|
|---|
| 309 | ($self->{asterisk}) && ($sz="q") ||
|
|---|
| 310 | ($mnemonic =~ /^v?mov([qd])$/) && ($sz=$1) ||
|
|---|
| 311 | ($mnemonic =~ /^v?pinsr([qdwb])$/) && ($sz=$1) ||
|
|---|
| 312 | ($mnemonic =~ /^vpbroadcast([qdwb])$/) && ($sz=$1) ||
|
|---|
| 313 | ($mnemonic =~ /^v(?!perm)[a-z]+[fi]128$/) && ($sz="x");
|
|---|
| 314 |
|
|---|
| 315 | $self->{opmask} =~ s/%(k[0-7])/$1/;
|
|---|
| 316 |
|
|---|
| 317 | if (defined($self->{index})) {
|
|---|
| 318 | sprintf "%s[%s%s*%d%s]%s",$szmap{$sz},
|
|---|
| 319 | $self->{label}?"$self->{label}+":"",
|
|---|
| 320 | $self->{index},$self->{scale},
|
|---|
| 321 | $self->{base}?"+$self->{base}":"",
|
|---|
| 322 | $self->{opmask};
|
|---|
| 323 | } elsif ($self->{base} eq "rip") {
|
|---|
| 324 | sprintf "%s[%s]",$szmap{$sz},$self->{label};
|
|---|
| 325 | } else {
|
|---|
| 326 | sprintf "%s[%s%s]%s", $szmap{$sz},
|
|---|
| 327 | $self->{label}?"$self->{label}+":"",
|
|---|
| 328 | $self->{base},$self->{opmask};
|
|---|
| 329 | }
|
|---|
| 330 | }
|
|---|
| 331 | }
|
|---|
| 332 | }
|
|---|
| 333 | { package register; # pick up registers, which start with %.
|
|---|
| 334 | sub re {
|
|---|
| 335 | my ($class, $line, $opcode) = @_;
|
|---|
| 336 | my $self = {};
|
|---|
| 337 | my $ret;
|
|---|
| 338 |
|
|---|
| 339 | # optional * ----vvv--- appears in indirect jmp/call
|
|---|
| 340 | if ($$line =~ /^(\*?)%(\w+)((?:{[^}]+})*)/) {
|
|---|
| 341 | bless $self,$class;
|
|---|
| 342 | $self->{asterisk} = $1;
|
|---|
| 343 | $self->{value} = $2;
|
|---|
| 344 | $self->{opmask} = $3;
|
|---|
| 345 | $opcode->size($self->size());
|
|---|
| 346 | $ret = $self;
|
|---|
| 347 | $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
|
|---|
| 348 | }
|
|---|
| 349 | $ret;
|
|---|
| 350 | }
|
|---|
| 351 | sub size {
|
|---|
| 352 | my $self = shift;
|
|---|
| 353 | my $ret;
|
|---|
| 354 |
|
|---|
| 355 | if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; }
|
|---|
| 356 | elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; }
|
|---|
| 357 | elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; }
|
|---|
| 358 | elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; }
|
|---|
| 359 | elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; }
|
|---|
| 360 | elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; }
|
|---|
| 361 | elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; }
|
|---|
| 362 | elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; }
|
|---|
| 363 |
|
|---|
| 364 | $ret;
|
|---|
| 365 | }
|
|---|
| 366 | sub out {
|
|---|
| 367 | my $self = shift;
|
|---|
| 368 | if ($gas) { sprintf "%s%%%s%s", $self->{asterisk},
|
|---|
| 369 | $self->{value},
|
|---|
| 370 | $self->{opmask}; }
|
|---|
| 371 | else { $self->{opmask} =~ s/%(k[0-7])/$1/;
|
|---|
| 372 | $self->{value}.$self->{opmask}; }
|
|---|
| 373 | }
|
|---|
| 374 | }
|
|---|
| 375 | { package label; # pick up labels, which end with :
|
|---|
| 376 | sub re {
|
|---|
| 377 | my ($class, $line) = @_;
|
|---|
| 378 | my $self = {};
|
|---|
| 379 | my $ret;
|
|---|
| 380 |
|
|---|
| 381 | if ($$line =~ /(^[\.\w]+)\:/) {
|
|---|
| 382 | bless $self,$class;
|
|---|
| 383 | $self->{value} = $1;
|
|---|
| 384 | $ret = $self;
|
|---|
| 385 | $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
|
|---|
| 386 |
|
|---|
| 387 | $self->{value} =~ s/^\.L/$decor/;
|
|---|
| 388 | }
|
|---|
| 389 | $ret;
|
|---|
| 390 | }
|
|---|
| 391 | sub out {
|
|---|
| 392 | my $self = shift;
|
|---|
| 393 |
|
|---|
| 394 | if ($gas) {
|
|---|
| 395 | my $func = ($globals{$self->{value}} or $self->{value}) . ":";
|
|---|
| 396 | if ($win64 && $current_function->{name} eq $self->{value}
|
|---|
| 397 | && $current_function->{abi} eq "svr4") {
|
|---|
| 398 | $func .= "\n";
|
|---|
| 399 | $func .= " movq %rdi,8(%rsp)\n";
|
|---|
| 400 | $func .= " movq %rsi,16(%rsp)\n";
|
|---|
| 401 | $func .= " movq %rsp,%rax\n";
|
|---|
| 402 | $func .= "${decor}SEH_begin_$current_function->{name}:\n";
|
|---|
| 403 | my $narg = $current_function->{narg};
|
|---|
| 404 | $narg=6 if (!defined($narg));
|
|---|
| 405 | $func .= " movq %rcx,%rdi\n" if ($narg>0);
|
|---|
| 406 | $func .= " movq %rdx,%rsi\n" if ($narg>1);
|
|---|
| 407 | $func .= " movq %r8,%rdx\n" if ($narg>2);
|
|---|
| 408 | $func .= " movq %r9,%rcx\n" if ($narg>3);
|
|---|
| 409 | $func .= " movq 40(%rsp),%r8\n" if ($narg>4);
|
|---|
| 410 | $func .= " movq 48(%rsp),%r9\n" if ($narg>5);
|
|---|
| 411 | }
|
|---|
| 412 | $func;
|
|---|
| 413 | } elsif ($self->{value} ne "$current_function->{name}") {
|
|---|
| 414 | # Make all labels in masm global.
|
|---|
| 415 | $self->{value} .= ":" if ($masm);
|
|---|
| 416 | $self->{value} . ":";
|
|---|
| 417 | } elsif ($win64 && $current_function->{abi} eq "svr4") {
|
|---|
| 418 | my $func = "$current_function->{name}" .
|
|---|
| 419 | ($nasm ? ":" : "\tPROC $current_function->{scope}") .
|
|---|
| 420 | "\n";
|
|---|
| 421 | $func .= " mov QWORD$PTR\[8+rsp\],rdi\t;WIN64 prologue\n";
|
|---|
| 422 | $func .= " mov QWORD$PTR\[16+rsp\],rsi\n";
|
|---|
| 423 | $func .= " mov rax,rsp\n";
|
|---|
| 424 | $func .= "${decor}SEH_begin_$current_function->{name}:";
|
|---|
| 425 | $func .= ":" if ($masm);
|
|---|
| 426 | $func .= "\n";
|
|---|
| 427 | my $narg = $current_function->{narg};
|
|---|
| 428 | $narg=6 if (!defined($narg));
|
|---|
| 429 | $func .= " mov rdi,rcx\n" if ($narg>0);
|
|---|
| 430 | $func .= " mov rsi,rdx\n" if ($narg>1);
|
|---|
| 431 | $func .= " mov rdx,r8\n" if ($narg>2);
|
|---|
| 432 | $func .= " mov rcx,r9\n" if ($narg>3);
|
|---|
| 433 | $func .= " mov r8,QWORD$PTR\[40+rsp\]\n" if ($narg>4);
|
|---|
| 434 | $func .= " mov r9,QWORD$PTR\[48+rsp\]\n" if ($narg>5);
|
|---|
| 435 | $func .= "\n";
|
|---|
| 436 | } else {
|
|---|
| 437 | "$current_function->{name}".
|
|---|
| 438 | ($nasm ? ":" : "\tPROC $current_function->{scope}");
|
|---|
| 439 | }
|
|---|
| 440 | }
|
|---|
| 441 | }
|
|---|
| 442 | { package expr; # pick up expressions
|
|---|
| 443 | sub re {
|
|---|
| 444 | my ($class, $line, $opcode) = @_;
|
|---|
| 445 | my $self = {};
|
|---|
| 446 | my $ret;
|
|---|
| 447 |
|
|---|
| 448 | if ($$line =~ /(^[^,]+)/) {
|
|---|
| 449 | bless $self,$class;
|
|---|
| 450 | $self->{value} = $1;
|
|---|
| 451 | $ret = $self;
|
|---|
| 452 | $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
|
|---|
| 453 |
|
|---|
| 454 | $self->{value} =~ s/\@PLT// if (!$elf);
|
|---|
| 455 | $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
|
|---|
| 456 | $self->{value} =~ s/\.L/$decor/g;
|
|---|
| 457 | $self->{opcode} = $opcode;
|
|---|
| 458 | }
|
|---|
| 459 | $ret;
|
|---|
| 460 | }
|
|---|
| 461 | sub out {
|
|---|
| 462 | my $self = shift;
|
|---|
| 463 | if ($nasm && $self->{opcode}->mnemonic()=~m/^j(?![re]cxz)/) {
|
|---|
| 464 | "NEAR ".$self->{value};
|
|---|
| 465 | } else {
|
|---|
| 466 | $self->{value};
|
|---|
| 467 | }
|
|---|
| 468 | }
|
|---|
| 469 | }
|
|---|
| 470 | { package cfi_directive;
|
|---|
| 471 | # CFI directives annotate instructions that are significant for
|
|---|
| 472 | # stack unwinding procedure compliant with DWARF specification,
|
|---|
| 473 | # see http://dwarfstd.org/. Besides naturally expected for this
|
|---|
| 474 | # script platform-specific filtering function, this module adds
|
|---|
| 475 | # three auxiliary synthetic directives not recognized by [GNU]
|
|---|
| 476 | # assembler:
|
|---|
| 477 | #
|
|---|
| 478 | # - .cfi_push to annotate push instructions in prologue, which
|
|---|
| 479 | # translates to .cfi_adjust_cfa_offset (if needed) and
|
|---|
| 480 | # .cfi_offset;
|
|---|
| 481 | # - .cfi_pop to annotate pop instructions in epilogue, which
|
|---|
| 482 | # translates to .cfi_adjust_cfa_offset (if needed) and
|
|---|
| 483 | # .cfi_restore;
|
|---|
| 484 | # - [and most notably] .cfi_cfa_expression which encodes
|
|---|
| 485 | # DW_CFA_def_cfa_expression and passes it to .cfi_escape as
|
|---|
| 486 | # byte vector;
|
|---|
| 487 | #
|
|---|
| 488 | # CFA expressions were introduced in DWARF specification version
|
|---|
| 489 | # 3 and describe how to deduce CFA, Canonical Frame Address. This
|
|---|
| 490 | # becomes handy if your stack frame is variable and you can't
|
|---|
| 491 | # spare register for [previous] frame pointer. Suggested directive
|
|---|
| 492 | # syntax is made-up mix of DWARF operator suffixes [subset of]
|
|---|
| 493 | # and references to registers with optional bias. Following example
|
|---|
| 494 | # describes offloaded *original* stack pointer at specific offset
|
|---|
| 495 | # from *current* stack pointer:
|
|---|
| 496 | #
|
|---|
| 497 | # .cfi_cfa_expression %rsp+40,deref,+8
|
|---|
| 498 | #
|
|---|
| 499 | # Final +8 has everything to do with the fact that CFA is defined
|
|---|
| 500 | # as reference to top of caller's stack, and on x86_64 call to
|
|---|
| 501 | # subroutine pushes 8-byte return address. In other words original
|
|---|
| 502 | # stack pointer upon entry to a subroutine is 8 bytes off from CFA.
|
|---|
| 503 |
|
|---|
| 504 | # Below constants are taken from "DWARF Expressions" section of the
|
|---|
| 505 | # DWARF specification, section is numbered 7.7 in versions 3 and 4.
|
|---|
| 506 | my %DW_OP_simple = ( # no-arg operators, mapped directly
|
|---|
| 507 | deref => 0x06, dup => 0x12,
|
|---|
| 508 | drop => 0x13, over => 0x14,
|
|---|
| 509 | pick => 0x15, swap => 0x16,
|
|---|
| 510 | rot => 0x17, xderef => 0x18,
|
|---|
| 511 |
|
|---|
| 512 | abs => 0x19, and => 0x1a,
|
|---|
| 513 | div => 0x1b, minus => 0x1c,
|
|---|
| 514 | mod => 0x1d, mul => 0x1e,
|
|---|
| 515 | neg => 0x1f, not => 0x20,
|
|---|
| 516 | or => 0x21, plus => 0x22,
|
|---|
| 517 | shl => 0x24, shr => 0x25,
|
|---|
| 518 | shra => 0x26, xor => 0x27,
|
|---|
| 519 | );
|
|---|
| 520 |
|
|---|
| 521 | my %DW_OP_complex = ( # used in specific subroutines
|
|---|
| 522 | constu => 0x10, # uleb128
|
|---|
| 523 | consts => 0x11, # sleb128
|
|---|
| 524 | plus_uconst => 0x23, # uleb128
|
|---|
| 525 | lit0 => 0x30, # add 0-31 to opcode
|
|---|
| 526 | reg0 => 0x50, # add 0-31 to opcode
|
|---|
| 527 | breg0 => 0x70, # add 0-31 to opcole, sleb128
|
|---|
| 528 | regx => 0x90, # uleb28
|
|---|
| 529 | fbreg => 0x91, # sleb128
|
|---|
| 530 | bregx => 0x92, # uleb128, sleb128
|
|---|
| 531 | piece => 0x93, # uleb128
|
|---|
| 532 | );
|
|---|
| 533 |
|
|---|
| 534 | # Following constants are defined in x86_64 ABI supplement, for
|
|---|
| 535 | # example available at https://www.uclibc.org/docs/psABI-x86_64.pdf,
|
|---|
| 536 | # see section 3.7 "Stack Unwind Algorithm".
|
|---|
| 537 | my %DW_reg_idx = (
|
|---|
| 538 | "%rax"=>0, "%rdx"=>1, "%rcx"=>2, "%rbx"=>3,
|
|---|
| 539 | "%rsi"=>4, "%rdi"=>5, "%rbp"=>6, "%rsp"=>7,
|
|---|
| 540 | "%r8" =>8, "%r9" =>9, "%r10"=>10, "%r11"=>11,
|
|---|
| 541 | "%r12"=>12, "%r13"=>13, "%r14"=>14, "%r15"=>15
|
|---|
| 542 | );
|
|---|
| 543 |
|
|---|
| 544 | my ($cfa_reg, $cfa_rsp);
|
|---|
| 545 | my @cfa_stack;
|
|---|
| 546 |
|
|---|
| 547 | # [us]leb128 format is variable-length integer representation base
|
|---|
| 548 | # 2^128, with most significant bit of each byte being 0 denoting
|
|---|
| 549 | # *last* most significant digit. See "Variable Length Data" in the
|
|---|
| 550 | # DWARF specification, numbered 7.6 at least in versions 3 and 4.
|
|---|
| 551 | sub sleb128 {
|
|---|
| 552 | use integer; # get right shift extend sign
|
|---|
| 553 |
|
|---|
| 554 | my $val = shift;
|
|---|
| 555 | my $sign = ($val < 0) ? -1 : 0;
|
|---|
| 556 | my @ret = ();
|
|---|
| 557 |
|
|---|
| 558 | while(1) {
|
|---|
| 559 | push @ret, $val&0x7f;
|
|---|
| 560 |
|
|---|
| 561 | # see if remaining bits are same and equal to most
|
|---|
| 562 | # significant bit of the current digit, if so, it's
|
|---|
| 563 | # last digit...
|
|---|
| 564 | last if (($val>>6) == $sign);
|
|---|
| 565 |
|
|---|
| 566 | @ret[-1] |= 0x80;
|
|---|
| 567 | $val >>= 7;
|
|---|
| 568 | }
|
|---|
| 569 |
|
|---|
| 570 | return @ret;
|
|---|
| 571 | }
|
|---|
| 572 | sub uleb128 {
|
|---|
| 573 | my $val = shift;
|
|---|
| 574 | my @ret = ();
|
|---|
| 575 |
|
|---|
| 576 | while(1) {
|
|---|
| 577 | push @ret, $val&0x7f;
|
|---|
| 578 |
|
|---|
| 579 | # see if it's last significant digit...
|
|---|
| 580 | last if (($val >>= 7) == 0);
|
|---|
| 581 |
|
|---|
| 582 | @ret[-1] |= 0x80;
|
|---|
| 583 | }
|
|---|
| 584 |
|
|---|
| 585 | return @ret;
|
|---|
| 586 | }
|
|---|
| 587 | sub const {
|
|---|
| 588 | my $val = shift;
|
|---|
| 589 |
|
|---|
| 590 | if ($val >= 0 && $val < 32) {
|
|---|
| 591 | return ($DW_OP_complex{lit0}+$val);
|
|---|
| 592 | }
|
|---|
| 593 | return ($DW_OP_complex{consts}, sleb128($val));
|
|---|
| 594 | }
|
|---|
| 595 | sub reg {
|
|---|
| 596 | my $val = shift;
|
|---|
| 597 |
|
|---|
| 598 | return if ($val !~ m/^(%r\w+)(?:([\+\-])((?:0x)?[0-9a-f]+))?/);
|
|---|
| 599 |
|
|---|
| 600 | my $reg = $DW_reg_idx{$1};
|
|---|
| 601 | my $off = eval ("0 $2 $3");
|
|---|
| 602 |
|
|---|
| 603 | return (($DW_OP_complex{breg0} + $reg), sleb128($off));
|
|---|
| 604 | # Yes, we use DW_OP_bregX+0 to push register value and not
|
|---|
| 605 | # DW_OP_regX, because latter would require even DW_OP_piece,
|
|---|
| 606 | # which would be a waste under the circumstances. If you have
|
|---|
| 607 | # to use DWP_OP_reg, use "regx:N"...
|
|---|
| 608 | }
|
|---|
| 609 | sub cfa_expression {
|
|---|
| 610 | my $line = shift;
|
|---|
| 611 | my @ret;
|
|---|
| 612 |
|
|---|
| 613 | foreach my $token (split(/,\s*/,$line)) {
|
|---|
| 614 | if ($token =~ /^%r/) {
|
|---|
| 615 | push @ret,reg($token);
|
|---|
| 616 | } elsif ($token =~ /((?:0x)?[0-9a-f]+)\((%r\w+)\)/) {
|
|---|
| 617 | push @ret,reg("$2+$1");
|
|---|
| 618 | } elsif ($token =~ /(\w+):(\-?(?:0x)?[0-9a-f]+)(U?)/i) {
|
|---|
| 619 | my $i = 1*eval($2);
|
|---|
| 620 | push @ret,$DW_OP_complex{$1}, ($3 ? uleb128($i) : sleb128($i));
|
|---|
| 621 | } elsif (my $i = 1*eval($token) or $token eq "0") {
|
|---|
| 622 | if ($token =~ /^\+/) {
|
|---|
| 623 | push @ret,$DW_OP_complex{plus_uconst},uleb128($i);
|
|---|
| 624 | } else {
|
|---|
| 625 | push @ret,const($i);
|
|---|
| 626 | }
|
|---|
| 627 | } else {
|
|---|
| 628 | push @ret,$DW_OP_simple{$token};
|
|---|
| 629 | }
|
|---|
| 630 | }
|
|---|
| 631 |
|
|---|
| 632 | # Finally we return DW_CFA_def_cfa_expression, 15, followed by
|
|---|
| 633 | # length of the expression and of course the expression itself.
|
|---|
| 634 | return (15,scalar(@ret),@ret);
|
|---|
| 635 | }
|
|---|
| 636 | sub re {
|
|---|
| 637 | my ($class, $line) = @_;
|
|---|
| 638 | my $self = {};
|
|---|
| 639 | my $ret;
|
|---|
| 640 |
|
|---|
| 641 | if ($$line =~ s/^\s*\.cfi_(\w+)\s*//) {
|
|---|
| 642 | bless $self,$class;
|
|---|
| 643 | $ret = $self;
|
|---|
| 644 | undef $self->{value};
|
|---|
| 645 | my $dir = $1;
|
|---|
| 646 |
|
|---|
| 647 | SWITCH: for ($dir) {
|
|---|
| 648 | # What is $cfa_rsp? Effectively it's difference between %rsp
|
|---|
| 649 | # value and current CFA, Canonical Frame Address, which is
|
|---|
| 650 | # why it starts with -8. Recall that CFA is top of caller's
|
|---|
| 651 | # stack...
|
|---|
| 652 | /startproc/ && do { ($cfa_reg, $cfa_rsp) = ("%rsp", -8); last; };
|
|---|
| 653 | /endproc/ && do { ($cfa_reg, $cfa_rsp) = ("%rsp", 0);
|
|---|
| 654 | # .cfi_remember_state directives that are not
|
|---|
| 655 | # matched with .cfi_restore_state are
|
|---|
| 656 | # unnecessary.
|
|---|
| 657 | die "unpaired .cfi_remember_state" if (@cfa_stack);
|
|---|
| 658 | last;
|
|---|
| 659 | };
|
|---|
| 660 | /def_cfa_register/
|
|---|
| 661 | && do { $cfa_reg = $$line; last; };
|
|---|
| 662 | /def_cfa_offset/
|
|---|
| 663 | && do { $cfa_rsp = -1*eval($$line) if ($cfa_reg eq "%rsp");
|
|---|
| 664 | last;
|
|---|
| 665 | };
|
|---|
| 666 | /adjust_cfa_offset/
|
|---|
| 667 | && do { $cfa_rsp -= 1*eval($$line) if ($cfa_reg eq "%rsp");
|
|---|
| 668 | last;
|
|---|
| 669 | };
|
|---|
| 670 | /def_cfa/ && do { if ($$line =~ /(%r\w+)\s*,\s*(.+)/) {
|
|---|
| 671 | $cfa_reg = $1;
|
|---|
| 672 | $cfa_rsp = -1*eval($2) if ($cfa_reg eq "%rsp");
|
|---|
| 673 | }
|
|---|
| 674 | last;
|
|---|
| 675 | };
|
|---|
| 676 | /push/ && do { $dir = undef;
|
|---|
| 677 | $cfa_rsp -= 8;
|
|---|
| 678 | if ($cfa_reg eq "%rsp") {
|
|---|
| 679 | $self->{value} = ".cfi_adjust_cfa_offset\t8\n";
|
|---|
| 680 | }
|
|---|
| 681 | $self->{value} .= ".cfi_offset\t$$line,$cfa_rsp";
|
|---|
| 682 | last;
|
|---|
| 683 | };
|
|---|
| 684 | /pop/ && do { $dir = undef;
|
|---|
| 685 | $cfa_rsp += 8;
|
|---|
| 686 | if ($cfa_reg eq "%rsp") {
|
|---|
| 687 | $self->{value} = ".cfi_adjust_cfa_offset\t-8\n";
|
|---|
| 688 | }
|
|---|
| 689 | $self->{value} .= ".cfi_restore\t$$line";
|
|---|
| 690 | last;
|
|---|
| 691 | };
|
|---|
| 692 | /cfa_expression/
|
|---|
| 693 | && do { $dir = undef;
|
|---|
| 694 | $self->{value} = ".cfi_escape\t" .
|
|---|
| 695 | join(",", map(sprintf("0x%02x", $_),
|
|---|
| 696 | cfa_expression($$line)));
|
|---|
| 697 | last;
|
|---|
| 698 | };
|
|---|
| 699 | /remember_state/
|
|---|
| 700 | && do { push @cfa_stack, [$cfa_reg, $cfa_rsp];
|
|---|
| 701 | last;
|
|---|
| 702 | };
|
|---|
| 703 | /restore_state/
|
|---|
| 704 | && do { ($cfa_reg, $cfa_rsp) = @{pop @cfa_stack};
|
|---|
| 705 | last;
|
|---|
| 706 | };
|
|---|
| 707 | }
|
|---|
| 708 |
|
|---|
| 709 | $self->{value} = ".cfi_$dir\t$$line" if ($dir);
|
|---|
| 710 |
|
|---|
| 711 | $$line = "";
|
|---|
| 712 | }
|
|---|
| 713 |
|
|---|
| 714 | return $ret;
|
|---|
| 715 | }
|
|---|
| 716 | sub out {
|
|---|
| 717 | my $self = shift;
|
|---|
| 718 | return ($elf ? $self->{value} : undef);
|
|---|
| 719 | }
|
|---|
| 720 | }
|
|---|
| 721 | { package directive; # pick up directives, which start with .
|
|---|
| 722 | sub re {
|
|---|
| 723 | my ($class, $line) = @_;
|
|---|
| 724 | my $self = {};
|
|---|
| 725 | my $ret;
|
|---|
| 726 | my $dir;
|
|---|
| 727 |
|
|---|
| 728 | # chain-call to cfi_directive
|
|---|
| 729 | $ret = cfi_directive->re($line) and return $ret;
|
|---|
| 730 |
|
|---|
| 731 | if ($$line =~ /^\s*(\.\w+)/) {
|
|---|
| 732 | bless $self,$class;
|
|---|
| 733 | $dir = $1;
|
|---|
| 734 | $ret = $self;
|
|---|
| 735 | undef $self->{value};
|
|---|
| 736 | $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
|
|---|
| 737 |
|
|---|
| 738 | SWITCH: for ($dir) {
|
|---|
| 739 | /\.global|\.globl|\.extern/
|
|---|
| 740 | && do { $globals{$$line} = $prefix . $$line;
|
|---|
| 741 | $$line = $globals{$$line} if ($prefix);
|
|---|
| 742 | last;
|
|---|
| 743 | };
|
|---|
| 744 | /\.type/ && do { my ($sym,$type,$narg) = split(',',$$line);
|
|---|
| 745 | if ($type eq "\@function") {
|
|---|
| 746 | undef $current_function;
|
|---|
| 747 | $current_function->{name} = $sym;
|
|---|
| 748 | $current_function->{abi} = "svr4";
|
|---|
| 749 | $current_function->{narg} = $narg;
|
|---|
| 750 | $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";
|
|---|
| 751 | } elsif ($type eq "\@abi-omnipotent") {
|
|---|
| 752 | undef $current_function;
|
|---|
| 753 | $current_function->{name} = $sym;
|
|---|
| 754 | $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";
|
|---|
| 755 | }
|
|---|
| 756 | $$line =~ s/\@abi\-omnipotent/\@function/;
|
|---|
| 757 | $$line =~ s/\@function.*/\@function/;
|
|---|
| 758 | last;
|
|---|
| 759 | };
|
|---|
| 760 | /\.asciz/ && do { if ($$line =~ /^"(.*)"$/) {
|
|---|
| 761 | $dir = ".byte";
|
|---|
| 762 | $$line = join(",",unpack("C*",$1),0);
|
|---|
| 763 | }
|
|---|
| 764 | last;
|
|---|
| 765 | };
|
|---|
| 766 | /\.rva|\.long|\.quad/
|
|---|
| 767 | && do { $$line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
|
|---|
| 768 | $$line =~ s/\.L/$decor/g;
|
|---|
| 769 | last;
|
|---|
| 770 | };
|
|---|
| 771 | }
|
|---|
| 772 |
|
|---|
| 773 | if ($gas) {
|
|---|
| 774 | $self->{value} = $dir . "\t" . $$line;
|
|---|
| 775 |
|
|---|
| 776 | if ($dir =~ /\.extern/) {
|
|---|
| 777 | $self->{value} = ""; # swallow extern
|
|---|
| 778 | } elsif (!$elf && $dir =~ /\.type/) {
|
|---|
| 779 | $self->{value} = "";
|
|---|
| 780 | $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" .
|
|---|
| 781 | (defined($globals{$1})?".scl 2;":".scl 3;") .
|
|---|
| 782 | "\t.type 32;\t.endef"
|
|---|
| 783 | if ($win64 && $$line =~ /([^,]+),\@function/);
|
|---|
| 784 | } elsif (!$elf && $dir =~ /\.size/) {
|
|---|
| 785 | $self->{value} = "";
|
|---|
| 786 | if (defined($current_function)) {
|
|---|
| 787 | $self->{value} .= "${decor}SEH_end_$current_function->{name}:"
|
|---|
| 788 | if ($win64 && $current_function->{abi} eq "svr4");
|
|---|
| 789 | undef $current_function;
|
|---|
| 790 | }
|
|---|
| 791 | } elsif (!$elf && $dir =~ /\.align/) {
|
|---|
| 792 | $self->{value} = ".p2align\t" . (log($$line)/log(2));
|
|---|
| 793 | } elsif ($dir eq ".section") {
|
|---|
| 794 | $current_segment=$$line;
|
|---|
| 795 | if (!$elf && $current_segment eq ".init") {
|
|---|
| 796 | if ($flavour eq "macosx") { $self->{value} = ".mod_init_func"; }
|
|---|
| 797 | elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.ctors"; }
|
|---|
| 798 | }
|
|---|
| 799 | } elsif ($dir =~ /\.(text|data)/) {
|
|---|
| 800 | $current_segment=".$1";
|
|---|
| 801 | } elsif ($dir =~ /\.hidden/) {
|
|---|
| 802 | if ($flavour eq "macosx") { $self->{value} = ".private_extern\t$prefix$$line"; }
|
|---|
| 803 | elsif ($flavour eq "mingw64") { $self->{value} = ""; }
|
|---|
| 804 | } elsif ($dir =~ /\.comm/) {
|
|---|
| 805 | $self->{value} = "$dir\t$prefix$$line";
|
|---|
| 806 | $self->{value} =~ s|,([0-9]+),([0-9]+)$|",$1,".log($2)/log(2)|e if ($flavour eq "macosx");
|
|---|
| 807 | }
|
|---|
| 808 | $$line = "";
|
|---|
| 809 | return $self;
|
|---|
| 810 | }
|
|---|
| 811 |
|
|---|
| 812 | # non-gas case or nasm/masm
|
|---|
| 813 | SWITCH: for ($dir) {
|
|---|
| 814 | /\.text/ && do { my $v=undef;
|
|---|
| 815 | if ($nasm) {
|
|---|
| 816 | $v="section .text code align=64\n";
|
|---|
| 817 | } else {
|
|---|
| 818 | $v="$current_segment\tENDS\n" if ($current_segment);
|
|---|
| 819 | $current_segment = ".text\$";
|
|---|
| 820 | $v.="$current_segment\tSEGMENT ";
|
|---|
| 821 | $v.=$masm>=$masmref ? "ALIGN(256)" : "PAGE";
|
|---|
| 822 | $v.=" 'CODE'";
|
|---|
| 823 | }
|
|---|
| 824 | $self->{value} = $v;
|
|---|
| 825 | last;
|
|---|
| 826 | };
|
|---|
| 827 | /\.data/ && do { my $v=undef;
|
|---|
| 828 | if ($nasm) {
|
|---|
| 829 | $v="section .data data align=8\n";
|
|---|
| 830 | } else {
|
|---|
| 831 | $v="$current_segment\tENDS\n" if ($current_segment);
|
|---|
| 832 | $current_segment = "_DATA";
|
|---|
| 833 | $v.="$current_segment\tSEGMENT";
|
|---|
| 834 | }
|
|---|
| 835 | $self->{value} = $v;
|
|---|
| 836 | last;
|
|---|
| 837 | };
|
|---|
| 838 | /\.section/ && do { my $v=undef;
|
|---|
| 839 | $$line =~ s/([^,]*).*/$1/;
|
|---|
| 840 | $$line = ".CRT\$XCU" if ($$line eq ".init");
|
|---|
| 841 | if ($nasm) {
|
|---|
| 842 | $v="section $$line";
|
|---|
| 843 | if ($$line=~/\.([px])data/) {
|
|---|
| 844 | $v.=" rdata align=";
|
|---|
| 845 | $v.=$1 eq "p"? 4 : 8;
|
|---|
| 846 | } elsif ($$line=~/\.CRT\$/i) {
|
|---|
| 847 | $v.=" rdata align=8";
|
|---|
| 848 | }
|
|---|
| 849 | } else {
|
|---|
| 850 | $v="$current_segment\tENDS\n" if ($current_segment);
|
|---|
| 851 | $v.="$$line\tSEGMENT";
|
|---|
| 852 | if ($$line=~/\.([px])data/) {
|
|---|
| 853 | $v.=" READONLY";
|
|---|
| 854 | $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref);
|
|---|
| 855 | } elsif ($$line=~/\.CRT\$/i) {
|
|---|
| 856 | $v.=" READONLY ";
|
|---|
| 857 | $v.=$masm>=$masmref ? "ALIGN(8)" : "DWORD";
|
|---|
| 858 | }
|
|---|
| 859 | }
|
|---|
| 860 | $current_segment = $$line;
|
|---|
| 861 | $self->{value} = $v;
|
|---|
| 862 | last;
|
|---|
| 863 | };
|
|---|
| 864 | /\.extern/ && do { $self->{value} = "EXTERN\t".$$line;
|
|---|
| 865 | $self->{value} .= ":NEAR" if ($masm);
|
|---|
| 866 | last;
|
|---|
| 867 | };
|
|---|
| 868 | /\.globl|.global/
|
|---|
| 869 | && do { $self->{value} = $masm?"PUBLIC":"global";
|
|---|
| 870 | $self->{value} .= "\t".$$line;
|
|---|
| 871 | last;
|
|---|
| 872 | };
|
|---|
| 873 | /\.size/ && do { if (defined($current_function)) {
|
|---|
| 874 | undef $self->{value};
|
|---|
| 875 | if ($current_function->{abi} eq "svr4") {
|
|---|
| 876 | $self->{value}="${decor}SEH_end_$current_function->{name}:";
|
|---|
| 877 | $self->{value}.=":\n" if($masm);
|
|---|
| 878 | }
|
|---|
| 879 | $self->{value}.="$current_function->{name}\tENDP" if($masm && $current_function->{name});
|
|---|
| 880 | undef $current_function;
|
|---|
| 881 | }
|
|---|
| 882 | last;
|
|---|
| 883 | };
|
|---|
| 884 | /\.align/ && do { my $max = ($masm && $masm>=$masmref) ? 256 : 4096;
|
|---|
| 885 | $self->{value} = "ALIGN\t".($$line>$max?$max:$$line);
|
|---|
| 886 | last;
|
|---|
| 887 | };
|
|---|
| 888 | /\.(value|long|rva|quad)/
|
|---|
| 889 | && do { my $sz = substr($1,0,1);
|
|---|
| 890 | my @arr = split(/,\s*/,$$line);
|
|---|
| 891 | my $last = pop(@arr);
|
|---|
| 892 | my $conv = sub { my $var=shift;
|
|---|
| 893 | $var=~s/^(0b[0-1]+)/oct($1)/eig;
|
|---|
| 894 | $var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm);
|
|---|
| 895 | if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva"))
|
|---|
| 896 | { $var=~s/^([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; }
|
|---|
| 897 | $var;
|
|---|
| 898 | };
|
|---|
| 899 |
|
|---|
| 900 | $sz =~ tr/bvlrq/BWDDQ/;
|
|---|
| 901 | $self->{value} = "\tD$sz\t";
|
|---|
| 902 | for (@arr) { $self->{value} .= &$conv($_).","; }
|
|---|
| 903 | $self->{value} .= &$conv($last);
|
|---|
| 904 | last;
|
|---|
| 905 | };
|
|---|
| 906 | /\.byte/ && do { my @str=split(/,\s*/,$$line);
|
|---|
| 907 | map(s/(0b[0-1]+)/oct($1)/eig,@str);
|
|---|
| 908 | map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm);
|
|---|
| 909 | while ($#str>15) {
|
|---|
| 910 | $self->{value}.="DB\t"
|
|---|
| 911 | .join(",",@str[0..15])."\n";
|
|---|
| 912 | foreach (0..15) { shift @str; }
|
|---|
| 913 | }
|
|---|
| 914 | $self->{value}.="DB\t"
|
|---|
| 915 | .join(",",@str) if (@str);
|
|---|
| 916 | last;
|
|---|
| 917 | };
|
|---|
| 918 | /\.comm/ && do { my @str=split(/,\s*/,$$line);
|
|---|
| 919 | my $v=undef;
|
|---|
| 920 | if ($nasm) {
|
|---|
| 921 | $v.="common $prefix@str[0] @str[1]";
|
|---|
| 922 | } else {
|
|---|
| 923 | $v="$current_segment\tENDS\n" if ($current_segment);
|
|---|
| 924 | $current_segment = "_DATA";
|
|---|
| 925 | $v.="$current_segment\tSEGMENT\n";
|
|---|
| 926 | $v.="COMM @str[0]:DWORD:".@str[1]/4;
|
|---|
| 927 | }
|
|---|
| 928 | $self->{value} = $v;
|
|---|
| 929 | last;
|
|---|
| 930 | };
|
|---|
| 931 | }
|
|---|
| 932 | $$line = "";
|
|---|
| 933 | }
|
|---|
| 934 |
|
|---|
| 935 | $ret;
|
|---|
| 936 | }
|
|---|
| 937 | sub out {
|
|---|
| 938 | my $self = shift;
|
|---|
| 939 | $self->{value};
|
|---|
| 940 | }
|
|---|
| 941 | }
|
|---|
| 942 |
|
|---|
| 943 | # Upon initial x86_64 introduction SSE>2 extensions were not introduced
|
|---|
| 944 | # yet. In order not to be bothered by tracing exact assembler versions,
|
|---|
| 945 | # but at the same time to provide a bare security minimum of AES-NI, we
|
|---|
| 946 | # hard-code some instructions. Extensions past AES-NI on the other hand
|
|---|
| 947 | # are traced by examining assembler version in individual perlasm
|
|---|
| 948 | # modules...
|
|---|
| 949 |
|
|---|
| 950 | my %regrm = ( "%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3,
|
|---|
| 951 | "%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7 );
|
|---|
| 952 |
|
|---|
| 953 | sub rex {
|
|---|
| 954 | my $opcode=shift;
|
|---|
| 955 | my ($dst,$src,$rex)=@_;
|
|---|
| 956 |
|
|---|
| 957 | $rex|=0x04 if($dst>=8);
|
|---|
| 958 | $rex|=0x01 if($src>=8);
|
|---|
| 959 | push @$opcode,($rex|0x40) if ($rex);
|
|---|
| 960 | }
|
|---|
| 961 |
|
|---|
| 962 | my $movq = sub { # elderly gas can't handle inter-register movq
|
|---|
| 963 | my $arg = shift;
|
|---|
| 964 | my @opcode=(0x66);
|
|---|
| 965 | if ($arg =~ /%xmm([0-9]+),\s*%r(\w+)/) {
|
|---|
| 966 | my ($src,$dst)=($1,$2);
|
|---|
| 967 | if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
|
|---|
| 968 | rex(\@opcode,$src,$dst,0x8);
|
|---|
| 969 | push @opcode,0x0f,0x7e;
|
|---|
| 970 | push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M
|
|---|
| 971 | @opcode;
|
|---|
| 972 | } elsif ($arg =~ /%r(\w+),\s*%xmm([0-9]+)/) {
|
|---|
| 973 | my ($src,$dst)=($2,$1);
|
|---|
| 974 | if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
|
|---|
| 975 | rex(\@opcode,$src,$dst,0x8);
|
|---|
| 976 | push @opcode,0x0f,0x6e;
|
|---|
| 977 | push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M
|
|---|
| 978 | @opcode;
|
|---|
| 979 | } else {
|
|---|
| 980 | ();
|
|---|
| 981 | }
|
|---|
| 982 | };
|
|---|
| 983 |
|
|---|
| 984 | my $pextrd = sub {
|
|---|
| 985 | if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*(%\w+)/) {
|
|---|
| 986 | my @opcode=(0x66);
|
|---|
| 987 | my $imm=$1;
|
|---|
| 988 | my $src=$2;
|
|---|
| 989 | my $dst=$3;
|
|---|
| 990 | if ($dst =~ /%r([0-9]+)d/) { $dst = $1; }
|
|---|
| 991 | elsif ($dst =~ /%e/) { $dst = $regrm{$dst}; }
|
|---|
| 992 | rex(\@opcode,$src,$dst);
|
|---|
| 993 | push @opcode,0x0f,0x3a,0x16;
|
|---|
| 994 | push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M
|
|---|
| 995 | push @opcode,$imm;
|
|---|
| 996 | @opcode;
|
|---|
| 997 | } else {
|
|---|
| 998 | ();
|
|---|
| 999 | }
|
|---|
| 1000 | };
|
|---|
| 1001 |
|
|---|
| 1002 | my $pinsrd = sub {
|
|---|
| 1003 | if (shift =~ /\$([0-9]+),\s*(%\w+),\s*%xmm([0-9]+)/) {
|
|---|
| 1004 | my @opcode=(0x66);
|
|---|
| 1005 | my $imm=$1;
|
|---|
| 1006 | my $src=$2;
|
|---|
| 1007 | my $dst=$3;
|
|---|
| 1008 | if ($src =~ /%r([0-9]+)/) { $src = $1; }
|
|---|
| 1009 | elsif ($src =~ /%e/) { $src = $regrm{$src}; }
|
|---|
| 1010 | rex(\@opcode,$dst,$src);
|
|---|
| 1011 | push @opcode,0x0f,0x3a,0x22;
|
|---|
| 1012 | push @opcode,0xc0|(($dst&7)<<3)|($src&7); # ModR/M
|
|---|
| 1013 | push @opcode,$imm;
|
|---|
| 1014 | @opcode;
|
|---|
| 1015 | } else {
|
|---|
| 1016 | ();
|
|---|
| 1017 | }
|
|---|
| 1018 | };
|
|---|
| 1019 |
|
|---|
| 1020 | my $pshufb = sub {
|
|---|
| 1021 | if (shift =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) {
|
|---|
| 1022 | my @opcode=(0x66);
|
|---|
| 1023 | rex(\@opcode,$2,$1);
|
|---|
| 1024 | push @opcode,0x0f,0x38,0x00;
|
|---|
| 1025 | push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M
|
|---|
| 1026 | @opcode;
|
|---|
| 1027 | } else {
|
|---|
| 1028 | ();
|
|---|
| 1029 | }
|
|---|
| 1030 | };
|
|---|
| 1031 |
|
|---|
| 1032 | my $palignr = sub {
|
|---|
| 1033 | if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
|
|---|
| 1034 | my @opcode=(0x66);
|
|---|
| 1035 | rex(\@opcode,$3,$2);
|
|---|
| 1036 | push @opcode,0x0f,0x3a,0x0f;
|
|---|
| 1037 | push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
|
|---|
| 1038 | push @opcode,$1;
|
|---|
| 1039 | @opcode;
|
|---|
| 1040 | } else {
|
|---|
| 1041 | ();
|
|---|
| 1042 | }
|
|---|
| 1043 | };
|
|---|
| 1044 |
|
|---|
| 1045 | my $pclmulqdq = sub {
|
|---|
| 1046 | if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
|
|---|
| 1047 | my @opcode=(0x66);
|
|---|
| 1048 | rex(\@opcode,$3,$2);
|
|---|
| 1049 | push @opcode,0x0f,0x3a,0x44;
|
|---|
| 1050 | push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
|
|---|
| 1051 | my $c=$1;
|
|---|
| 1052 | push @opcode,$c=~/^0/?oct($c):$c;
|
|---|
| 1053 | @opcode;
|
|---|
| 1054 | } else {
|
|---|
| 1055 | ();
|
|---|
| 1056 | }
|
|---|
| 1057 | };
|
|---|
| 1058 |
|
|---|
| 1059 | my $rdrand = sub {
|
|---|
| 1060 | if (shift =~ /%[er](\w+)/) {
|
|---|
| 1061 | my @opcode=();
|
|---|
| 1062 | my $dst=$1;
|
|---|
| 1063 | if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
|
|---|
| 1064 | rex(\@opcode,0,$dst,8);
|
|---|
| 1065 | push @opcode,0x0f,0xc7,0xf0|($dst&7);
|
|---|
| 1066 | @opcode;
|
|---|
| 1067 | } else {
|
|---|
| 1068 | ();
|
|---|
| 1069 | }
|
|---|
| 1070 | };
|
|---|
| 1071 |
|
|---|
| 1072 | my $rdseed = sub {
|
|---|
| 1073 | if (shift =~ /%[er](\w+)/) {
|
|---|
| 1074 | my @opcode=();
|
|---|
| 1075 | my $dst=$1;
|
|---|
| 1076 | if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
|
|---|
| 1077 | rex(\@opcode,0,$dst,8);
|
|---|
| 1078 | push @opcode,0x0f,0xc7,0xf8|($dst&7);
|
|---|
| 1079 | @opcode;
|
|---|
| 1080 | } else {
|
|---|
| 1081 | ();
|
|---|
| 1082 | }
|
|---|
| 1083 | };
|
|---|
| 1084 |
|
|---|
| 1085 | # Not all AVX-capable assemblers recognize AMD XOP extension. Since we
|
|---|
| 1086 | # are using only two instructions hand-code them in order to be excused
|
|---|
| 1087 | # from chasing assembler versions...
|
|---|
| 1088 |
|
|---|
| 1089 | sub rxb {
|
|---|
| 1090 | my $opcode=shift;
|
|---|
| 1091 | my ($dst,$src1,$src2,$rxb)=@_;
|
|---|
| 1092 |
|
|---|
| 1093 | $rxb|=0x7<<5;
|
|---|
| 1094 | $rxb&=~(0x04<<5) if($dst>=8);
|
|---|
| 1095 | $rxb&=~(0x01<<5) if($src1>=8);
|
|---|
| 1096 | $rxb&=~(0x02<<5) if($src2>=8);
|
|---|
| 1097 | push @$opcode,$rxb;
|
|---|
| 1098 | }
|
|---|
| 1099 |
|
|---|
| 1100 | my $vprotd = sub {
|
|---|
| 1101 | if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
|
|---|
| 1102 | my @opcode=(0x8f);
|
|---|
| 1103 | rxb(\@opcode,$3,$2,-1,0x08);
|
|---|
| 1104 | push @opcode,0x78,0xc2;
|
|---|
| 1105 | push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
|
|---|
| 1106 | my $c=$1;
|
|---|
| 1107 | push @opcode,$c=~/^0/?oct($c):$c;
|
|---|
| 1108 | @opcode;
|
|---|
| 1109 | } else {
|
|---|
| 1110 | ();
|
|---|
| 1111 | }
|
|---|
| 1112 | };
|
|---|
| 1113 |
|
|---|
| 1114 | my $vprotq = sub {
|
|---|
| 1115 | if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
|
|---|
| 1116 | my @opcode=(0x8f);
|
|---|
| 1117 | rxb(\@opcode,$3,$2,-1,0x08);
|
|---|
| 1118 | push @opcode,0x78,0xc3;
|
|---|
| 1119 | push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
|
|---|
| 1120 | my $c=$1;
|
|---|
| 1121 | push @opcode,$c=~/^0/?oct($c):$c;
|
|---|
| 1122 | @opcode;
|
|---|
| 1123 | } else {
|
|---|
| 1124 | ();
|
|---|
| 1125 | }
|
|---|
| 1126 | };
|
|---|
| 1127 |
|
|---|
| 1128 | # Intel Control-flow Enforcement Technology extension. All functions and
|
|---|
| 1129 | # indirect branch targets will have to start with this instruction...
|
|---|
| 1130 |
|
|---|
| 1131 | my $endbranch = sub {
|
|---|
| 1132 | (0xf3,0x0f,0x1e,0xfa);
|
|---|
| 1133 | };
|
|---|
| 1134 |
|
|---|
| 1135 | ########################################################################
|
|---|
| 1136 |
|
|---|
| 1137 | if ($nasm) {
|
|---|
| 1138 | print <<___;
|
|---|
| 1139 | default rel
|
|---|
| 1140 | %define XMMWORD
|
|---|
| 1141 | %define YMMWORD
|
|---|
| 1142 | %define ZMMWORD
|
|---|
| 1143 | ___
|
|---|
| 1144 | } elsif ($masm) {
|
|---|
| 1145 | print <<___;
|
|---|
| 1146 | OPTION DOTNAME
|
|---|
| 1147 | ___
|
|---|
| 1148 | }
|
|---|
| 1149 | while(defined(my $line=<>)) {
|
|---|
| 1150 |
|
|---|
| 1151 | $line =~ s|\R$||; # Better chomp
|
|---|
| 1152 |
|
|---|
| 1153 | $line =~ s|[#!].*$||; # get rid of asm-style comments...
|
|---|
| 1154 | $line =~ s|/\*.*\*/||; # ... and C-style comments...
|
|---|
| 1155 | $line =~ s|^\s+||; # ... and skip white spaces in beginning
|
|---|
| 1156 | $line =~ s|\s+$||; # ... and at the end
|
|---|
| 1157 |
|
|---|
| 1158 | if (my $label=label->re(\$line)) { print $label->out(); }
|
|---|
| 1159 |
|
|---|
| 1160 | if (my $directive=directive->re(\$line)) {
|
|---|
| 1161 | printf "%s",$directive->out();
|
|---|
| 1162 | } elsif (my $opcode=opcode->re(\$line)) {
|
|---|
| 1163 | my $asm = eval("\$".$opcode->mnemonic());
|
|---|
| 1164 |
|
|---|
| 1165 | if ((ref($asm) eq 'CODE') && scalar(my @bytes=&$asm($line))) {
|
|---|
| 1166 | print $gas?".byte\t":"DB\t",join(',',@bytes),"\n";
|
|---|
| 1167 | next;
|
|---|
| 1168 | }
|
|---|
| 1169 |
|
|---|
| 1170 | my @args;
|
|---|
| 1171 | ARGUMENT: while (1) {
|
|---|
| 1172 | my $arg;
|
|---|
| 1173 |
|
|---|
| 1174 | ($arg=register->re(\$line, $opcode))||
|
|---|
| 1175 | ($arg=const->re(\$line)) ||
|
|---|
| 1176 | ($arg=ea->re(\$line, $opcode)) ||
|
|---|
| 1177 | ($arg=expr->re(\$line, $opcode)) ||
|
|---|
| 1178 | last ARGUMENT;
|
|---|
| 1179 |
|
|---|
| 1180 | push @args,$arg;
|
|---|
| 1181 |
|
|---|
| 1182 | last ARGUMENT if ($line !~ /^,/);
|
|---|
| 1183 |
|
|---|
| 1184 | $line =~ s/^,\s*//;
|
|---|
| 1185 | } # ARGUMENT:
|
|---|
| 1186 |
|
|---|
| 1187 | if ($#args>=0) {
|
|---|
| 1188 | my $insn;
|
|---|
| 1189 | my $sz=$opcode->size();
|
|---|
| 1190 |
|
|---|
| 1191 | if ($gas) {
|
|---|
| 1192 | $insn = $opcode->out($#args>=1?$args[$#args]->size():$sz);
|
|---|
| 1193 | @args = map($_->out($sz),@args);
|
|---|
| 1194 | printf "\t%s\t%s",$insn,join(",",@args);
|
|---|
| 1195 | } else {
|
|---|
| 1196 | $insn = $opcode->out();
|
|---|
| 1197 | foreach (@args) {
|
|---|
| 1198 | my $arg = $_->out();
|
|---|
| 1199 | # $insn.=$sz compensates for movq, pinsrw, ...
|
|---|
| 1200 | if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; }
|
|---|
| 1201 | if ($arg =~ /^ymm[0-9]+$/) { $insn.=$sz; $sz="y" if(!$sz); last; }
|
|---|
| 1202 | if ($arg =~ /^zmm[0-9]+$/) { $insn.=$sz; $sz="z" if(!$sz); last; }
|
|---|
| 1203 | if ($arg =~ /^mm[0-9]+$/) { $insn.=$sz; $sz="q" if(!$sz); last; }
|
|---|
| 1204 | }
|
|---|
| 1205 | @args = reverse(@args);
|
|---|
| 1206 | undef $sz if ($nasm && $opcode->mnemonic() eq "lea");
|
|---|
| 1207 | printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args));
|
|---|
| 1208 | }
|
|---|
| 1209 | } else {
|
|---|
| 1210 | printf "\t%s",$opcode->out();
|
|---|
| 1211 | }
|
|---|
| 1212 | }
|
|---|
| 1213 |
|
|---|
| 1214 | print $line,"\n";
|
|---|
| 1215 | }
|
|---|
| 1216 |
|
|---|
| 1217 | print "\n$current_segment\tENDS\n" if ($current_segment && $masm);
|
|---|
| 1218 | print "END\n" if ($masm);
|
|---|
| 1219 |
|
|---|
| 1220 | close STDOUT or die "error closing STDOUT: $!";
|
|---|
| 1221 |
|
|---|
| 1222 | |
|---|
| 1223 | #################################################
|
|---|
| 1224 | # Cross-reference x86_64 ABI "card"
|
|---|
| 1225 | #
|
|---|
| 1226 | # Unix Win64
|
|---|
| 1227 | # %rax * *
|
|---|
| 1228 | # %rbx - -
|
|---|
| 1229 | # %rcx #4 #1
|
|---|
| 1230 | # %rdx #3 #2
|
|---|
| 1231 | # %rsi #2 -
|
|---|
| 1232 | # %rdi #1 -
|
|---|
| 1233 | # %rbp - -
|
|---|
| 1234 | # %rsp - -
|
|---|
| 1235 | # %r8 #5 #3
|
|---|
| 1236 | # %r9 #6 #4
|
|---|
| 1237 | # %r10 * *
|
|---|
| 1238 | # %r11 * *
|
|---|
| 1239 | # %r12 - -
|
|---|
| 1240 | # %r13 - -
|
|---|
| 1241 | # %r14 - -
|
|---|
| 1242 | # %r15 - -
|
|---|
| 1243 | #
|
|---|
| 1244 | # (*) volatile register
|
|---|
| 1245 | # (-) preserved by callee
|
|---|
| 1246 | # (#) Nth argument, volatile
|
|---|
| 1247 | #
|
|---|
| 1248 | # In Unix terms top of stack is argument transfer area for arguments
|
|---|
| 1249 | # which could not be accommodated in registers. Or in other words 7th
|
|---|
| 1250 | # [integer] argument resides at 8(%rsp) upon function entry point.
|
|---|
| 1251 | # 128 bytes above %rsp constitute a "red zone" which is not touched
|
|---|
| 1252 | # by signal handlers and can be used as temporal storage without
|
|---|
| 1253 | # allocating a frame.
|
|---|
| 1254 | #
|
|---|
| 1255 | # In Win64 terms N*8 bytes on top of stack is argument transfer area,
|
|---|
| 1256 | # which belongs to/can be overwritten by callee. N is the number of
|
|---|
| 1257 | # arguments passed to callee, *but* not less than 4! This means that
|
|---|
| 1258 | # upon function entry point 5th argument resides at 40(%rsp), as well
|
|---|
| 1259 | # as that 32 bytes from 8(%rsp) can always be used as temporal
|
|---|
| 1260 | # storage [without allocating a frame]. One can actually argue that
|
|---|
| 1261 | # one can assume a "red zone" above stack pointer under Win64 as well.
|
|---|
| 1262 | # Point is that at apparently no occasion Windows kernel would alter
|
|---|
| 1263 | # the area above user stack pointer in true asynchronous manner...
|
|---|
| 1264 | #
|
|---|
| 1265 | # All the above means that if assembler programmer adheres to Unix
|
|---|
| 1266 | # register and stack layout, but disregards the "red zone" existence,
|
|---|
| 1267 | # it's possible to use following prologue and epilogue to "gear" from
|
|---|
| 1268 | # Unix to Win64 ABI in leaf functions with not more than 6 arguments.
|
|---|
| 1269 | #
|
|---|
| 1270 | # omnipotent_function:
|
|---|
| 1271 | # ifdef WIN64
|
|---|
| 1272 | # movq %rdi,8(%rsp)
|
|---|
| 1273 | # movq %rsi,16(%rsp)
|
|---|
| 1274 | # movq %rcx,%rdi ; if 1st argument is actually present
|
|---|
| 1275 | # movq %rdx,%rsi ; if 2nd argument is actually ...
|
|---|
| 1276 | # movq %r8,%rdx ; if 3rd argument is ...
|
|---|
| 1277 | # movq %r9,%rcx ; if 4th argument ...
|
|---|
| 1278 | # movq 40(%rsp),%r8 ; if 5th ...
|
|---|
| 1279 | # movq 48(%rsp),%r9 ; if 6th ...
|
|---|
| 1280 | # endif
|
|---|
| 1281 | # ...
|
|---|
| 1282 | # ifdef WIN64
|
|---|
| 1283 | # movq 8(%rsp),%rdi
|
|---|
| 1284 | # movq 16(%rsp),%rsi
|
|---|
| 1285 | # endif
|
|---|
| 1286 | # ret
|
|---|
| 1287 | #
|
|---|
| 1288 | |
|---|
| 1289 | #################################################
|
|---|
| 1290 | # Win64 SEH, Structured Exception Handling.
|
|---|
| 1291 | #
|
|---|
| 1292 | # Unlike on Unix systems(*) lack of Win64 stack unwinding information
|
|---|
| 1293 | # has undesired side-effect at run-time: if an exception is raised in
|
|---|
| 1294 | # assembler subroutine such as those in question (basically we're
|
|---|
| 1295 | # referring to segmentation violations caused by malformed input
|
|---|
| 1296 | # parameters), the application is briskly terminated without invoking
|
|---|
| 1297 | # any exception handlers, most notably without generating memory dump
|
|---|
| 1298 | # or any user notification whatsoever. This poses a problem. It's
|
|---|
| 1299 | # possible to address it by registering custom language-specific
|
|---|
| 1300 | # handler that would restore processor context to the state at
|
|---|
| 1301 | # subroutine entry point and return "exception is not handled, keep
|
|---|
| 1302 | # unwinding" code. Writing such handler can be a challenge... But it's
|
|---|
| 1303 | # doable, though requires certain coding convention. Consider following
|
|---|
| 1304 | # snippet:
|
|---|
| 1305 | #
|
|---|
| 1306 | # .type function,@function
|
|---|
| 1307 | # function:
|
|---|
| 1308 | # movq %rsp,%rax # copy rsp to volatile register
|
|---|
| 1309 | # pushq %r15 # save non-volatile registers
|
|---|
| 1310 | # pushq %rbx
|
|---|
| 1311 | # pushq %rbp
|
|---|
| 1312 | # movq %rsp,%r11
|
|---|
| 1313 | # subq %rdi,%r11 # prepare [variable] stack frame
|
|---|
| 1314 | # andq $-64,%r11
|
|---|
| 1315 | # movq %rax,0(%r11) # check for exceptions
|
|---|
| 1316 | # movq %r11,%rsp # allocate [variable] stack frame
|
|---|
| 1317 | # movq %rax,0(%rsp) # save original rsp value
|
|---|
| 1318 | # magic_point:
|
|---|
| 1319 | # ...
|
|---|
| 1320 | # movq 0(%rsp),%rcx # pull original rsp value
|
|---|
| 1321 | # movq -24(%rcx),%rbp # restore non-volatile registers
|
|---|
| 1322 | # movq -16(%rcx),%rbx
|
|---|
| 1323 | # movq -8(%rcx),%r15
|
|---|
| 1324 | # movq %rcx,%rsp # restore original rsp
|
|---|
| 1325 | # magic_epilogue:
|
|---|
| 1326 | # ret
|
|---|
| 1327 | # .size function,.-function
|
|---|
| 1328 | #
|
|---|
| 1329 | # The key is that up to magic_point copy of original rsp value remains
|
|---|
| 1330 | # in chosen volatile register and no non-volatile register, except for
|
|---|
| 1331 | # rsp, is modified. While past magic_point rsp remains constant till
|
|---|
| 1332 | # the very end of the function. In this case custom language-specific
|
|---|
| 1333 | # exception handler would look like this:
|
|---|
| 1334 | #
|
|---|
| 1335 | # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
|
|---|
| 1336 | # CONTEXT *context,DISPATCHER_CONTEXT *disp)
|
|---|
| 1337 | # { ULONG64 *rsp = (ULONG64 *)context->Rax;
|
|---|
| 1338 | # ULONG64 rip = context->Rip;
|
|---|
| 1339 | #
|
|---|
| 1340 | # if (rip >= magic_point)
|
|---|
| 1341 | # { rsp = (ULONG64 *)context->Rsp;
|
|---|
| 1342 | # if (rip < magic_epilogue)
|
|---|
| 1343 | # { rsp = (ULONG64 *)rsp[0];
|
|---|
| 1344 | # context->Rbp = rsp[-3];
|
|---|
| 1345 | # context->Rbx = rsp[-2];
|
|---|
| 1346 | # context->R15 = rsp[-1];
|
|---|
| 1347 | # }
|
|---|
| 1348 | # }
|
|---|
| 1349 | # context->Rsp = (ULONG64)rsp;
|
|---|
| 1350 | # context->Rdi = rsp[1];
|
|---|
| 1351 | # context->Rsi = rsp[2];
|
|---|
| 1352 | #
|
|---|
| 1353 | # memcpy (disp->ContextRecord,context,sizeof(CONTEXT));
|
|---|
| 1354 | # RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase,
|
|---|
| 1355 | # dips->ControlPc,disp->FunctionEntry,disp->ContextRecord,
|
|---|
| 1356 | # &disp->HandlerData,&disp->EstablisherFrame,NULL);
|
|---|
| 1357 | # return ExceptionContinueSearch;
|
|---|
| 1358 | # }
|
|---|
| 1359 | #
|
|---|
| 1360 | # It's appropriate to implement this handler in assembler, directly in
|
|---|
| 1361 | # function's module. In order to do that one has to know members'
|
|---|
| 1362 | # offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant
|
|---|
| 1363 | # values. Here they are:
|
|---|
| 1364 | #
|
|---|
| 1365 | # CONTEXT.Rax 120
|
|---|
| 1366 | # CONTEXT.Rcx 128
|
|---|
| 1367 | # CONTEXT.Rdx 136
|
|---|
| 1368 | # CONTEXT.Rbx 144
|
|---|
| 1369 | # CONTEXT.Rsp 152
|
|---|
| 1370 | # CONTEXT.Rbp 160
|
|---|
| 1371 | # CONTEXT.Rsi 168
|
|---|
| 1372 | # CONTEXT.Rdi 176
|
|---|
| 1373 | # CONTEXT.R8 184
|
|---|
| 1374 | # CONTEXT.R9 192
|
|---|
| 1375 | # CONTEXT.R10 200
|
|---|
| 1376 | # CONTEXT.R11 208
|
|---|
| 1377 | # CONTEXT.R12 216
|
|---|
| 1378 | # CONTEXT.R13 224
|
|---|
| 1379 | # CONTEXT.R14 232
|
|---|
| 1380 | # CONTEXT.R15 240
|
|---|
| 1381 | # CONTEXT.Rip 248
|
|---|
| 1382 | # CONTEXT.Xmm6 512
|
|---|
| 1383 | # sizeof(CONTEXT) 1232
|
|---|
| 1384 | # DISPATCHER_CONTEXT.ControlPc 0
|
|---|
| 1385 | # DISPATCHER_CONTEXT.ImageBase 8
|
|---|
| 1386 | # DISPATCHER_CONTEXT.FunctionEntry 16
|
|---|
| 1387 | # DISPATCHER_CONTEXT.EstablisherFrame 24
|
|---|
| 1388 | # DISPATCHER_CONTEXT.TargetIp 32
|
|---|
| 1389 | # DISPATCHER_CONTEXT.ContextRecord 40
|
|---|
| 1390 | # DISPATCHER_CONTEXT.LanguageHandler 48
|
|---|
| 1391 | # DISPATCHER_CONTEXT.HandlerData 56
|
|---|
| 1392 | # UNW_FLAG_NHANDLER 0
|
|---|
| 1393 | # ExceptionContinueSearch 1
|
|---|
| 1394 | #
|
|---|
| 1395 | # In order to tie the handler to the function one has to compose
|
|---|
| 1396 | # couple of structures: one for .xdata segment and one for .pdata.
|
|---|
| 1397 | #
|
|---|
| 1398 | # UNWIND_INFO structure for .xdata segment would be
|
|---|
| 1399 | #
|
|---|
| 1400 | # function_unwind_info:
|
|---|
| 1401 | # .byte 9,0,0,0
|
|---|
| 1402 | # .rva handler
|
|---|
| 1403 | #
|
|---|
| 1404 | # This structure designates exception handler for a function with
|
|---|
| 1405 | # zero-length prologue, no stack frame or frame register.
|
|---|
| 1406 | #
|
|---|
| 1407 | # To facilitate composing of .pdata structures, auto-generated "gear"
|
|---|
| 1408 | # prologue copies rsp value to rax and denotes next instruction with
|
|---|
| 1409 | # .LSEH_begin_{function_name} label. This essentially defines the SEH
|
|---|
| 1410 | # styling rule mentioned in the beginning. Position of this label is
|
|---|
| 1411 | # chosen in such manner that possible exceptions raised in the "gear"
|
|---|
| 1412 | # prologue would be accounted to caller and unwound from latter's frame.
|
|---|
| 1413 | # End of function is marked with respective .LSEH_end_{function_name}
|
|---|
| 1414 | # label. To summarize, .pdata segment would contain
|
|---|
| 1415 | #
|
|---|
| 1416 | # .rva .LSEH_begin_function
|
|---|
| 1417 | # .rva .LSEH_end_function
|
|---|
| 1418 | # .rva function_unwind_info
|
|---|
| 1419 | #
|
|---|
| 1420 | # Reference to function_unwind_info from .xdata segment is the anchor.
|
|---|
| 1421 | # In case you wonder why references are 32-bit .rvas and not 64-bit
|
|---|
| 1422 | # .quads. References put into these two segments are required to be
|
|---|
| 1423 | # *relative* to the base address of the current binary module, a.k.a.
|
|---|
| 1424 | # image base. No Win64 module, be it .exe or .dll, can be larger than
|
|---|
| 1425 | # 2GB and thus such relative references can be and are accommodated in
|
|---|
| 1426 | # 32 bits.
|
|---|
| 1427 | #
|
|---|
| 1428 | # Having reviewed the example function code, one can argue that "movq
|
|---|
| 1429 | # %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix
|
|---|
| 1430 | # rax would contain an undefined value. If this "offends" you, use
|
|---|
| 1431 | # another register and refrain from modifying rax till magic_point is
|
|---|
| 1432 | # reached, i.e. as if it was a non-volatile register. If more registers
|
|---|
| 1433 | # are required prior [variable] frame setup is completed, note that
|
|---|
| 1434 | # nobody says that you can have only one "magic point." You can
|
|---|
| 1435 | # "liberate" non-volatile registers by denoting last stack off-load
|
|---|
| 1436 | # instruction and reflecting it in finer grade unwind logic in handler.
|
|---|
| 1437 | # After all, isn't it why it's called *language-specific* handler...
|
|---|
| 1438 | #
|
|---|
| 1439 | # SE handlers are also involved in unwinding stack when executable is
|
|---|
| 1440 | # profiled or debugged. Profiling implies additional limitations that
|
|---|
| 1441 | # are too subtle to discuss here. For now it's sufficient to say that
|
|---|
| 1442 | # in order to simplify handlers one should either a) offload original
|
|---|
| 1443 | # %rsp to stack (like discussed above); or b) if you have a register to
|
|---|
| 1444 | # spare for frame pointer, choose volatile one.
|
|---|
| 1445 | #
|
|---|
| 1446 | # (*) Note that we're talking about run-time, not debug-time. Lack of
|
|---|
| 1447 | # unwind information makes debugging hard on both Windows and
|
|---|
| 1448 | # Unix. "Unlike" refers to the fact that on Unix signal handler
|
|---|
| 1449 | # will always be invoked, core dumped and appropriate exit code
|
|---|
| 1450 | # returned to parent (for user notification).
|
|---|
