1 #!/usr/bin/env perl 2 3 # Ascetic x86_64 AT&T to MASM/NASM assembler translator by <appro>. 4 # 5 # Why AT&T to MASM and not vice versa? Several reasons. Because AT&T 6 # format is way easier to parse. Because it's simpler to "gear" from 7 # Unix ABI to Windows one [see cross-reference "card" at the end of 8 # file]. Because Linux targets were available first... 9 # 10 # In addition the script also "distills" code suitable for GNU 11 # assembler, so that it can be compiled with more rigid assemblers, 12 # such as Solaris /usr/ccs/bin/as. 13 # 14 # This translator is not designed to convert *arbitrary* assembler 15 # code from AT&T format to MASM one. It's designed to convert just 16 # enough to provide for dual-ABI OpenSSL modules development... 17 # There *are* limitations and you might have to modify your assembler 18 # code or this script to achieve the desired result... 19 # 20 # Currently recognized limitations: 21 # 22 # - can't use multiple ops per line; 23 # 24 # Dual-ABI styling rules. 25 # 26 # 1. Adhere to Unix register and stack layout [see cross-reference 27 # ABI "card" at the end for explanation]. 28 # 2. Forget about "red zone," stick to more traditional blended 29 # stack frame allocation. If volatile storage is actually required 30 # that is. If not, just leave the stack as is. 31 # 3. Functions tagged with ".type name,@function" get crafted with 32 # unified Win64 prologue and epilogue automatically. If you want 33 # to take care of ABI differences yourself, tag functions as 34 # ".type name,@abi-omnipotent" instead. 35 # 4. To optimize the Win64 prologue you can specify number of input 36 # arguments as ".type name,@function,N." Keep in mind that if N is 37 # larger than 6, then you *have to* write "abi-omnipotent" code, 38 # because >6 cases can't be addressed with unified prologue. 39 # 5. Name local labels as .L*, do *not* use dynamic labels such as 1: 40 # (sorry about latter). 41 # 6. Don't use [or hand-code with .byte] "rep ret." "ret" mnemonic is 42 # required to identify the spots, where to inject Win64 epilogue! 43 # But on the pros, it's then prefixed with rep automatically:-) 44 # 7. Stick to explicit ip-relative addressing. If you have to use 45 # GOTPCREL addressing, stick to mov symbol@GOTPCREL(%rip),%r??. 46 # Both are recognized and translated to proper Win64 addressing 47 # modes. To support legacy code a synthetic directive, .picmeup, 48 # is implemented. It puts address of the *next* instruction into 49 # target register, e.g.: 50 # 51 # .picmeup %rax 52 # lea .Label-.(%rax),%rax 53 # 54 # 8. In order to provide for structured exception handling unified 55 # Win64 prologue copies %rsp value to %rax. For further details 56 # see SEH paragraph at the end. 57 # 9. .init segment is allowed to contain calls to functions only. 58 # a. If function accepts more than 4 arguments *and* >4th argument 59 # is declared as non 64-bit value, do clear its upper part. 60 62 my $flavour = shift; 63 my $output = shift; 64 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; } 65 66 open STDOUT,">$output" || die "can't open $output: $!" 67 if (defined($output)); 68 69 my $gas=1; $gas=0 if ($output =~ /\.asm$/); 70 my $elf=1; $elf=0 if (!$gas); 71 my $win64=0; 72 my $prefix=""; 73 my $decor=".L"; 74 75 my $masmref=8 + 50727*2**-32; # 8.00.50727 shipped with VS2005 76 my $masm=0; 77 my $PTR=" PTR"; 78 79 my $nasmref=2.03; 80 my $nasm=0; 81 82 if ($flavour eq "mingw64") { $gas=1; $elf=0; $win64=1; 83 $prefix=`echo __USER_LABEL_PREFIX__ | $ENV{CC} -E -P -`; 84 chomp($prefix); 85 } 86 elsif ($flavour eq "macosx") { $gas=1; $elf=0; $prefix="_"; $decor="L\$"; } 87 elsif ($flavour eq "masm") { $gas=0; $elf=0; $masm=$masmref; $win64=1; $decor="\$L\$"; } 88 elsif ($flavour eq "nasm") { $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $decor="\$L\$"; $PTR=""; } 89 elsif (!$gas) 90 { if ($ENV{ASM} =~ m/nasm/ && `nasm -v` =~ m/version ([0-9]+)\.([0-9]+)/i) 91 { $nasm = $1 + $2*0.01; $PTR=""; } 92 elsif (`ml64 2>&1` =~ m/Version ([0-9]+)\.([0-9]+)(\.([0-9]+))?/) 93 { $masm = $1 + $2*2**-16 + $4*2**-32; } 94 die "no assembler found on %PATH" if (!($nasm || $masm)); 95 $win64=1; 96 $elf=0; 97 $decor="\$L\$"; 98 } 99 100 my $current_segment; 101 my $current_function; 102 my %globals; 103 104 { package opcode; # pick up opcodes 105 sub re { 106 my $self = shift; # single instance in enough... 107 local *line = shift; 108 undef $ret; 109 110 if ($line =~ /^([a-z][a-z0-9]*)/i) { 111 $self->{op} = $1; 112 $ret = $self; 113 $line = substr($line,@+[0]); $line =~ s/^\s+//; 114 115 undef $self->{sz}; 116 if ($self->{op} =~ /^(movz)x?([bw]).*/) { # movz is pain... 117 $self->{op} = $1; 118 $self->{sz} = $2; 119 } elsif ($self->{op} =~ /call|jmp/) { 120 $self->{sz} = ""; 121 } elsif ($self->{op} =~ /^p/ && $' !~ /^(ush|op|insrw)/) { # SSEn 122 $self->{sz} = ""; 123 } elsif ($self->{op} =~ /^v/) { # VEX 124 $self->{sz} = ""; 125 } elsif ($self->{op} =~ /movq/ && $line =~ /%xmm/) { 126 $self->{sz} = ""; 127 } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) { 128 $self->{op} = $1; 129 $self->{sz} = $2; 130 } 131 } 132 $ret; 133 } 134 sub size { 135 my $self = shift; 136 my $sz = shift; 137 $self->{sz} = $sz if (defined($sz) && !defined($self->{sz})); 138 $self->{sz}; 139 } 140 sub out { 141 my $self = shift; 142 if ($gas) { 143 if ($self->{op} eq "movz") { # movz is pain... 144 sprintf "%s%s%s",$self->{op},$self->{sz},shift; 145 } elsif ($self->{op} =~ /^set/) { 146 "$self->{op}"; 147 } elsif ($self->{op} eq "ret") { 148 my $epilogue = ""; 149 if ($win64 && $current_function->{abi} eq "svr4") { 150 $epilogue = "movq 8(%rsp),%rdi\n\t" . 151 "movq 16(%rsp),%rsi\n\t"; 152 } 153 $epilogue . ".byte 0xf3,0xc3"; 154 } elsif ($self->{op} eq "call" && !$elf && $current_segment eq ".init") { 155 ".p2align\t3\n\t.quad"; 156 } else { 157 "$self->{op}$self->{sz}"; 158 } 159 } else { 160 $self->{op} =~ s/^movz/movzx/; 161 if ($self->{op} eq "ret") { 162 $self->{op} = ""; 163 if ($win64 && $current_function->{abi} eq "svr4") { 164 $self->{op} = "mov rdi,QWORD${PTR}[8+rsp]\t;WIN64 epilogue\n\t". 165 "mov rsi,QWORD${PTR}[16+rsp]\n\t"; 166 } 167 $self->{op} .= "DB\t0F3h,0C3h\t\t;repret"; 168 } elsif ($self->{op} =~ /^(pop|push)f/) { 169 $self->{op} .= $self->{sz}; 170 } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") { 171 $self->{op} = "\tDQ"; 172 } 173 $self->{op}; 174 } 175 } 176 sub mnemonic { 177 my $self=shift; 178 my $op=shift; 179 $self->{op}=$op if (defined($op)); 180 $self->{op}; 181 } 182 } 183 { package const; # pick up constants, which start with $ 184 sub re { 185 my $self = shift; # single instance in enough... 186 local *line = shift; 187 undef $ret; 188 189 if ($line =~ /^\$([^,]+)/) { 190 $self->{value} = $1; 191 $ret = $self; 192 $line = substr($line,@+[0]); $line =~ s/^\s+//; 193 } 194 $ret; 195 } 196 sub out { 197 my $self = shift; 198 199 if ($gas) { 200 # Solaris /usr/ccs/bin/as can't handle multiplications 201 # in $self->{value} 202 $self->{value} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi; 203 $self->{value} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg; 204 sprintf "\$%s",$self->{value}; 205 } else { 206 $self->{value} =~ s/(0b[0-1]+)/oct($1)/eig; 207 $self->{value} =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm); 208 sprintf "%s",$self->{value}; 209 } 210 } 211 } 212 { package ea; # pick up effective addresses: expr(%reg,%reg,scale) 213 sub re { 214 my $self = shift; # single instance in enough... 215 local *line = shift; 216 undef $ret; 217 218 # optional * ---vvv--- appears in indirect jmp/call 219 if ($line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)/) { 220 $self->{asterisk} = $1; 221 $self->{label} = $2; 222 ($self->{base},$self->{index},$self->{scale})=split(/,/,$3); 223 $self->{scale} = 1 if (!defined($self->{scale})); 224 $ret = $self; 225 $line = substr($line,@+[0]); $line =~ s/^\s+//; 226 227 if ($win64 && $self->{label} =~ s/\@GOTPCREL//) { 228 die if (opcode->mnemonic() ne "mov"); 229 opcode->mnemonic("lea"); 230 } 231 $self->{base} =~ s/^%//; 232 $self->{index} =~ s/^%// if (defined($self->{index})); 233 } 234 $ret; 235 } 236 sub size {} 237 sub out { 238 my $self = shift; 239 my $sz = shift; 240 241 $self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 242 $self->{label} =~ s/\.L/$decor/g; 243 244 # Silently convert all EAs to 64-bit. This is required for 245 # elder GNU assembler and results in more compact code, 246 # *but* most importantly AES module depends on this feature! 247 $self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; 248 $self->{base} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; 249 250 # Solaris /usr/ccs/bin/as can't handle multiplications 251 # in $self->{label}, new gas requires sign extension... 252 use integer; 253 $self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi; 254 $self->{label} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg; 255 $self->{label} =~ s/([0-9]+)/$1<<32>>32/eg; 256 257 if ($gas) { 258 $self->{label} =~ s/^___imp_/__imp__/ if ($flavour eq "mingw64"); 259 260 if (defined($self->{index})) { 261 sprintf "%s%s(%s,%%%s,%d)",$self->{asterisk}, 262 $self->{label}, 263 $self->{base}?"%$self->{base}":"", 264 $self->{index},$self->{scale}; 265 } else { 266 sprintf "%s%s(%%%s)", $self->{asterisk},$self->{label},$self->{base}; 267 } 268 } else { 269 %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR", l=>"DWORD$PTR", 270 q=>"QWORD$PTR",o=>"OWORD$PTR",x=>"XMMWORD$PTR" ); 271 272 $self->{label} =~ s/\./\$/g; 273 $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig; 274 $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/); 275 $sz="q" if ($self->{asterisk} || opcode->mnemonic() eq "movq"); 276 $sz="l" if (opcode->mnemonic() eq "movd"); 277 278 if (defined($self->{index})) { 279 sprintf "%s[%s%s*%d%s]",$szmap{$sz}, 280 $self->{label}?"$self->{label}+":"", 281 $self->{index},$self->{scale}, 282 $self->{base}?"+$self->{base}":""; 283 } elsif ($self->{base} eq "rip") { 284 sprintf "%s[%s]",$szmap{$sz},$self->{label}; 285 } else { 286 sprintf "%s[%s%s]",$szmap{$sz}, 287 $self->{label}?"$self->{label}+":"", 288 $self->{base}; 289 } 290 } 291 } 292 } 293 { package register; # pick up registers, which start with %. 294 sub re { 295 my $class = shift; # muliple instances... 296 my $self = {}; 297 local *line = shift; 298 undef $ret; 299 300 # optional * ---vvv--- appears in indirect jmp/call 301 if ($line =~ /^(\*?)%(\w+)/) { 302 bless $self,$class; 303 $self->{asterisk} = $1; 304 $self->{value} = $2; 305 $ret = $self; 306 $line = substr($line,@+[0]); $line =~ s/^\s+//; 307 } 308 $ret; 309 } 310 sub size { 311 my $self = shift; 312 undef $ret; 313 314 if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; } 315 elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; } 316 elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; } 317 elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; } 318 elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; } 319 elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; } 320 elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; } 321 elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; } 322 323 $ret; 324 } 325 sub out { 326 my $self = shift; 327 if ($gas) { sprintf "%s%%%s",$self->{asterisk},$self->{value}; } 328 else { $self->{value}; } 329 } 330 } 331 { package label; # pick up labels, which end with : 332 sub re { 333 my $self = shift; # single instance is enough... 334 local *line = shift; 335 undef $ret; 336 337 if ($line =~ /(^[\.\w]+)\:/) { 338 $self->{value} = $1; 339 $ret = $self; 340 $line = substr($line,@+[0]); $line =~ s/^\s+//; 341 342 $self->{value} =~ s/^\.L/$decor/; 343 } 344 $ret; 345 } 346 sub out { 347 my $self = shift; 348 349 if ($gas) { 350 my $func = ($globals{$self->{value}} or $self->{value}) . ":"; 351 if ($win64 && 352 $current_function->{name} eq $self->{value} && 353 $current_function->{abi} eq "svr4") { 354 $func .= "\n"; 355 $func .= " movq %rdi,8(%rsp)\n"; 356 $func .= " movq %rsi,16(%rsp)\n"; 357 $func .= " movq %rsp,%rax\n"; 358 $func .= "${decor}SEH_begin_$current_function->{name}:\n"; 359 my $narg = $current_function->{narg}; 360 $narg=6 if (!defined($narg)); 361 $func .= " movq %rcx,%rdi\n" if ($narg>0); 362 $func .= " movq %rdx,%rsi\n" if ($narg>1); 363 $func .= " movq %r8,%rdx\n" if ($narg>2); 364 $func .= " movq %r9,%rcx\n" if ($narg>3); 365 $func .= " movq 40(%rsp),%r8\n" if ($narg>4); 366 $func .= " movq 48(%rsp),%r9\n" if ($narg>5); 367 } 368 $func; 369 } elsif ($self->{value} ne "$current_function->{name}") { 370 $self->{value} .= ":" if ($masm && $ret!~m/^\$/); 371 $self->{value} . ":"; 372 } elsif ($win64 && $current_function->{abi} eq "svr4") { 373 my $func = "$current_function->{name}" . 374 ($nasm ? ":" : "\tPROC $current_function->{scope}") . 375 "\n"; 376 $func .= " mov QWORD${PTR}[8+rsp],rdi\t;WIN64 prologue\n"; 377 $func .= " mov QWORD${PTR}[16+rsp],rsi\n"; 378 $func .= " mov rax,rsp\n"; 379 $func .= "${decor}SEH_begin_$current_function->{name}:"; 380 $func .= ":" if ($masm); 381 $func .= "\n"; 382 my $narg = $current_function->{narg}; 383 $narg=6 if (!defined($narg)); 384 $func .= " mov rdi,rcx\n" if ($narg>0); 385 $func .= " mov rsi,rdx\n" if ($narg>1); 386 $func .= " mov rdx,r8\n" if ($narg>2); 387 $func .= " mov rcx,r9\n" if ($narg>3); 388 $func .= " mov r8,QWORD${PTR}[40+rsp]\n" if ($narg>4); 389 $func .= " mov r9,QWORD${PTR}[48+rsp]\n" if ($narg>5); 390 $func .= "\n"; 391 } else { 392 "$current_function->{name}". 393 ($nasm ? ":" : "\tPROC $current_function->{scope}"); 394 } 395 } 396 } 397 { package expr; # pick up expressioins 398 sub re { 399 my $self = shift; # single instance is enough... 400 local *line = shift; 401 undef $ret; 402 403 if ($line =~ /(^[^,]+)/) { 404 $self->{value} = $1; 405 $ret = $self; 406 $line = substr($line,@+[0]); $line =~ s/^\s+//; 407 408 $self->{value} =~ s/\@PLT// if (!$elf); 409 $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 410 $self->{value} =~ s/\.L/$decor/g; 411 } 412 $ret; 413 } 414 sub out { 415 my $self = shift; 416 if ($nasm && opcode->mnemonic()=~m/^j/) { 417 "NEAR ".$self->{value}; 418 } else { 419 $self->{value}; 420 } 421 } 422 } 423 { package directive; # pick up directives, which start with . 424 sub re { 425 my $self = shift; # single instance is enough... 426 local *line = shift; 427 undef $ret; 428 my $dir; 429 my %opcode = # lea 2f-1f(%rip),%dst; 1: nop; 2: 430 ( "%rax"=>0x01058d48, "%rcx"=>0x010d8d48, 431 "%rdx"=>0x01158d48, "%rbx"=>0x011d8d48, 432 "%rsp"=>0x01258d48, "%rbp"=>0x012d8d48, 433 "%rsi"=>0x01358d48, "%rdi"=>0x013d8d48, 434 "%r8" =>0x01058d4c, "%r9" =>0x010d8d4c, 435 "%r10"=>0x01158d4c, "%r11"=>0x011d8d4c, 436 "%r12"=>0x01258d4c, "%r13"=>0x012d8d4c, 437 "%r14"=>0x01358d4c, "%r15"=>0x013d8d4c ); 438 439 if ($line =~ /^\s*(\.\w+)/) { 440 $dir = $1; 441 $ret = $self; 442 undef $self->{value}; 443 $line = substr($line,@+[0]); $line =~ s/^\s+//; 444 445 SWITCH: for ($dir) { 446 /\.picmeup/ && do { if ($line =~ /(%r[\w]+)/i) { 447 $dir="\t.long"; 448 $line=sprintf "0x%x,0x90000000",$opcode{$1}; 449 } 450 last; 451 }; 452 /\.global|\.globl|\.extern/ 453 && do { $globals{$line} = $prefix . $line; 454 $line = $globals{$line} if ($prefix); 455 last; 456 }; 457 /\.type/ && do { ($sym,$type,$narg) = split(',',$line); 458 if ($type eq "\@function") { 459 undef $current_function; 460 $current_function->{name} = $sym; 461 $current_function->{abi} = "svr4"; 462 $current_function->{narg} = $narg; 463 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 464 } elsif ($type eq "\@abi-omnipotent") { 465 undef $current_function; 466 $current_function->{name} = $sym; 467 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 468 } 469 $line =~ s/\@abi\-omnipotent/\@function/; 470 $line =~ s/\@function.*/\@function/; 471 last; 472 }; 473 /\.asciz/ && do { if ($line =~ /^"(.*)"$/) { 474 $dir = ".byte"; 475 $line = join(",",unpack("C*",$1),0); 476 } 477 last; 478 }; 479 /\.rva|\.long|\.quad/ 480 && do { $line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 481 $line =~ s/\.L/$decor/g; 482 last; 483 }; 484 } 485 486 if ($gas) { 487 $self->{value} = $dir . "\t" . $line; 488 489 if ($dir =~ /\.extern/) { 490 $self->{value} = ""; # swallow extern 491 } elsif (!$elf && $dir =~ /\.type/) { 492 $self->{value} = ""; 493 $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" . 494 (defined($globals{$1})?".scl 2;":".scl 3;") . 495 "\t.type 32;\t.endef" 496 if ($win64 && $line =~ /([^,]+),\@function/); 497 } elsif (!$elf && $dir =~ /\.size/) { 498 $self->{value} = ""; 499 if (defined($current_function)) { 500 $self->{value} .= "${decor}SEH_end_$current_function->{name}:" 501 if ($win64 && $current_function->{abi} eq "svr4"); 502 undef $current_function; 503 } 504 } elsif (!$elf && $dir =~ /\.align/) { 505 $self->{value} = ".p2align\t" . (log($line)/log(2)); 506 } elsif ($dir eq ".section") { 507 $current_segment=$line; 508 if (!$elf && $current_segment eq ".init") { 509 if ($flavour eq "macosx") { $self->{value} = ".mod_init_func"; } 510 elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.ctors"; } 511 } 512 } elsif ($dir =~ /\.(text|data)/) { 513 $current_segment=".$1"; 514 } elsif ($dir =~ /\.hidden/) { 515 if ($flavour eq "macosx") { $self->{value} = ".private_extern\t$prefix$line"; } 516 elsif ($flavour eq "mingw64") { $self->{value} = ""; } 517 } elsif ($dir =~ /\.comm/) { 518 $self->{value} = "$dir\t$prefix$line"; 519 $self->{value} =~ s|,([0-9]+),([0-9]+)$|",$1,".log($2)/log(2)|e if ($flavour eq "macosx"); 520 } 521 $line = ""; 522 return $self; 523 } 524 525 # non-gas case or nasm/masm 526 SWITCH: for ($dir) { 527 /\.text/ && do { my $v=undef; 528 if ($nasm) { 529 $v="section .text code align=64\n"; 530 } else { 531 $v="$current_segment\tENDS\n" if ($current_segment); 532 $current_segment = ".text\$"; 533 $v.="$current_segment\tSEGMENT "; 534 $v.=$masm>=$masmref ? "ALIGN(64)" : "PAGE"; 535 $v.=" 'CODE'"; 536 } 537 $self->{value} = $v; 538 last; 539 }; 540 /\.data/ && do { my $v=undef; 541 if ($nasm) { 542 $v="section .data data align=8\n"; 543 } else { 544 $v="$current_segment\tENDS\n" if ($current_segment); 545 $current_segment = "_DATA"; 546 $v.="$current_segment\tSEGMENT"; 547 } 548 $self->{value} = $v; 549 last; 550 }; 551 /\.section/ && do { my $v=undef; 552 $line =~ s/([^,]*).*/$1/; 553 $line = ".CRT\$XCU" if ($line eq ".init"); 554 if ($nasm) { 555 $v="section $line"; 556 if ($line=~/\.([px])data/) { 557 $v.=" rdata align="; 558 $v.=$1 eq "p"? 4 : 8; 559 } elsif ($line=~/\.CRT\$/i) { 560 $v.=" rdata align=8"; 561 } 562 } else { 563 $v="$current_segment\tENDS\n" if ($current_segment); 564 $v.="$line\tSEGMENT"; 565 if ($line=~/\.([px])data/) { 566 $v.=" READONLY"; 567 $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref); 568 } elsif ($line=~/\.CRT\$/i) { 569 $v.=" READONLY "; 570 $v.=$masm>=$masmref ? "ALIGN(8)" : "DWORD"; 571 } 572 } 573 $current_segment = $line; 574 $self->{value} = $v; 575 last; 576 }; 577 /\.extern/ && do { $self->{value} = "EXTERN\t".$line; 578 $self->{value} .= ":NEAR" if ($masm); 579 last; 580 }; 581 /\.globl|.global/ 582 && do { $self->{value} = $masm?"PUBLIC":"global"; 583 $self->{value} .= "\t".$line; 584 last; 585 }; 586 /\.size/ && do { if (defined($current_function)) { 587 undef $self->{value}; 588 if ($current_function->{abi} eq "svr4") { 589 $self->{value}="${decor}SEH_end_$current_function->{name}:"; 590 $self->{value}.=":\n" if($masm); 591 } 592 $self->{value}.="$current_function->{name}\tENDP" if($masm && $current_function->{name}); 593 undef $current_function; 594 } 595 last; 596 }; 597 /\.align/ && do { $self->{value} = "ALIGN\t".$line; last; }; 598 /\.(value|long|rva|quad)/ 599 && do { my $sz = substr($1,0,1); 600 my @arr = split(/,\s*/,$line); 601 my $last = pop(@arr); 602 my $conv = sub { my $var=shift; 603 $var=~s/^(0b[0-1]+)/oct($1)/eig; 604 $var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm); 605 if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva")) 606 { $var=~s/([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; } 607 $var; 608 }; 609 610 $sz =~ tr/bvlrq/BWDDQ/; 611 $self->{value} = "\tD$sz\t"; 612 for (@arr) { $self->{value} .= &$conv($_).","; } 613 $self->{value} .= &$conv($last); 614 last; 615 }; 616 /\.byte/ && do { my @str=split(/,\s*/,$line); 617 map(s/(0b[0-1]+)/oct($1)/eig,@str); 618 map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm); 619 while ($#str>15) { 620 $self->{value}.="DB\t" 621 .join(",",@str[0..15])."\n"; 622 foreach (0..15) { shift @str; } 623 } 624 $self->{value}.="DB\t" 625 .join(",",@str) if (@str); 626 last; 627 }; 628 /\.comm/ && do { my @str=split(/,\s*/,$line); 629 my $v=undef; 630 if ($nasm) { 631 $v.="common $prefix@str[0] @str[1]"; 632 } else { 633 $v="$current_segment\tENDS\n" if ($current_segment); 634 $current_segment = "_DATA"; 635 $v.="$current_segment\tSEGMENT\n"; 636 $v.="COMM @str[0]:DWORD:".@str[1]/4; 637 } 638 $self->{value} = $v; 639 last; 640 }; 641 } 642 $line = ""; 643 } 644 645 $ret; 646 } 647 sub out { 648 my $self = shift; 649 $self->{value}; 650 } 651 } 652 653 sub rex { 654 local *opcode=shift; 655 my ($dst,$src,$rex)=@_; 656 657 $rex|=0x04 if($dst>=8); 658 $rex|=0x01 if($src>=8); 659 push @opcode,($rex|0x40) if ($rex); 660 } 661 662 # older gas and ml64 don't handle SSE>2 instructions 663 my %regrm = ( "%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3, 664 "%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7 ); 665 666 my $movq = sub { # elderly gas can't handle inter-register movq 667 my $arg = shift; 668 my @opcode=(0x66); 669 if ($arg =~ /%xmm([0-9]+),\s*%r(\w+)/) { 670 my ($src,$dst)=($1,$2); 671 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 672 rex(\@opcode,$src,$dst,0x8); 673 push @opcode,0x0f,0x7e; 674 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 675 @opcode; 676 } elsif ($arg =~ /%r(\w+),\s*%xmm([0-9]+)/) { 677 my ($src,$dst)=($2,$1); 678 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 679 rex(\@opcode,$src,$dst,0x8); 680 push @opcode,0x0f,0x6e; 681 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 682 @opcode; 683 } else { 684 (); 685 } 686 }; 687 688 my $pextrd = sub { 689 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*(%\w+)/) { 690 my @opcode=(0x66); 691 $imm=$1; 692 $src=$2; 693 $dst=$3; 694 if ($dst =~ /%r([0-9]+)d/) { $dst = $1; } 695 elsif ($dst =~ /%e/) { $dst = $regrm{$dst}; } 696 rex(\@opcode,$src,$dst); 697 push @opcode,0x0f,0x3a,0x16; 698 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 699 push @opcode,$imm; 700 @opcode; 701 } else { 702 (); 703 } 704 }; 705 706 my $pinsrd = sub { 707 if (shift =~ /\$([0-9]+),\s*(%\w+),\s*%xmm([0-9]+)/) { 708 my @opcode=(0x66); 709 $imm=$1; 710 $src=$2; 711 $dst=$3; 712 if ($src =~ /%r([0-9]+)/) { $src = $1; } 713 elsif ($src =~ /%e/) { $src = $regrm{$src}; } 714 rex(\@opcode,$dst,$src); 715 push @opcode,0x0f,0x3a,0x22; 716 push @opcode,0xc0|(($dst&7)<<3)|($src&7); # ModR/M 717 push @opcode,$imm; 718 @opcode; 719 } else { 720 (); 721 } 722 }; 723 724 my $pshufb = sub { 725 if (shift =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) { 726 my @opcode=(0x66); 727 rex(\@opcode,$2,$1); 728 push @opcode,0x0f,0x38,0x00; 729 push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M 730 @opcode; 731 } else { 732 (); 733 } 734 }; 735 736 my $palignr = sub { 737 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 738 my @opcode=(0x66); 739 rex(\@opcode,$3,$2); 740 push @opcode,0x0f,0x3a,0x0f; 741 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 742 push @opcode,$1; 743 @opcode; 744 } else { 745 (); 746 } 747 }; 748 749 my $pclmulqdq = sub { 750 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 751 my @opcode=(0x66); 752 rex(\@opcode,$3,$2); 753 push @opcode,0x0f,0x3a,0x44; 754 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 755 my $c=$1; 756 push @opcode,$c=~/^0/?oct($c):$c; 757 @opcode; 758 } else { 759 (); 760 } 761 }; 762 763 my $rdrand = sub { 764 if (shift =~ /%[er](\w+)/) { 765 my @opcode=(); 766 my $dst=$1; 767 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 768 rex(\@opcode,0,$1,8); 769 push @opcode,0x0f,0xc7,0xf0|($dst&7); 770 @opcode; 771 } else { 772 (); 773 } 774 }; 775 776 if ($nasm) { 777 print <<___; 778 default rel 779 %define XMMWORD 780 ___ 781 } elsif ($masm) { 782 print <<___; 783 OPTION DOTNAME 784 ___ 785 } 786 while($line=<>) { 787 788 chomp($line); 789 790 $line =~ s|[#!].*$||; # get rid of asm-style comments... 791 $line =~ s|/\*.*\*/||; # ... and C-style comments... 792 $line =~ s|^\s+||; # ... and skip white spaces in beginning 793 794 undef $label; 795 undef $opcode; 796 undef @args; 797 798 if ($label=label->re(\$line)) { print $label->out(); } 799 800 if (directive->re(\$line)) { 801 printf "%s",directive->out(); 802 } elsif ($opcode=opcode->re(\$line)) { 803 my $asm = eval("\$".$opcode->mnemonic()); 804 undef @bytes; 805 806 if ((ref($asm) eq 'CODE') && scalar(@bytes=&$asm($line))) { 807 print $gas?".byte\t":"DB\t",join(',',@bytes),"\n"; 808 next; 809 } 810 811 ARGUMENT: while (1) { 812 my $arg; 813 814 if ($arg=register->re(\$line)) { opcode->size($arg->size()); } 815 elsif ($arg=const->re(\$line)) { } 816 elsif ($arg=ea->re(\$line)) { } 817 elsif ($arg=expr->re(\$line)) { } 818 else { last ARGUMENT; } 819 820 push @args,$arg; 821 822 last ARGUMENT if ($line !~ /^,/); 823 824 $line =~ s/^,\s*//; 825 } # ARGUMENT: 826 827 if ($#args>=0) { 828 my $insn; 829 my $sz=opcode->size(); 830 831 if ($gas) { 832 $insn = $opcode->out($#args>=1?$args[$#args]->size():$sz); 833 @args = map($_->out($sz),@args); 834 printf "\t%s\t%s",$insn,join(",",@args); 835 } else { 836 $insn = $opcode->out(); 837 foreach (@args) { 838 my $arg = $_->out(); 839 # $insn.=$sz compensates for movq, pinsrw, ... 840 if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; } 841 if ($arg =~ /^mm[0-9]+$/) { $insn.=$sz; $sz="q" if(!$sz); last; } 842 } 843 @args = reverse(@args); 844 undef $sz if ($nasm && $opcode->mnemonic() eq "lea"); 845 printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args)); 846 } 847 } else { 848 printf "\t%s",$opcode->out(); 849 } 850 } 851 852 print $line,"\n"; 853 } 854 855 print "\n$current_segment\tENDS\n" if ($current_segment && $masm); 856 print "END\n" if ($masm); 857 858 close STDOUT; 859 860 ################################################# 862 # Cross-reference x86_64 ABI "card" 863 # 864 # Unix Win64 865 # %rax * * 866 # %rbx - - 867 # %rcx #4 #1 868 # %rdx #3 #2 869 # %rsi #2 - 870 # %rdi #1 - 871 # %rbp - - 872 # %rsp - - 873 # %r8 #5 #3 874 # %r9 #6 #4 875 # %r10 * * 876 # %r11 * * 877 # %r12 - - 878 # %r13 - - 879 # %r14 - - 880 # %r15 - - 881 # 882 # (*) volatile register 883 # (-) preserved by callee 884 # (#) Nth argument, volatile 885 # 886 # In Unix terms top of stack is argument transfer area for arguments 887 # which could not be accomodated in registers. Or in other words 7th 888 # [integer] argument resides at 8(%rsp) upon function entry point. 889 # 128 bytes above %rsp constitute a "red zone" which is not touched 890 # by signal handlers and can be used as temporal storage without 891 # allocating a frame. 892 # 893 # In Win64 terms N*8 bytes on top of stack is argument transfer area, 894 # which belongs to/can be overwritten by callee. N is the number of 895 # arguments passed to callee, *but* not less than 4! This means that 896 # upon function entry point 5th argument resides at 40(%rsp), as well 897 # as that 32 bytes from 8(%rsp) can always be used as temporal 898 # storage [without allocating a frame]. One can actually argue that 899 # one can assume a "red zone" above stack pointer under Win64 as well. 900 # Point is that at apparently no occasion Windows kernel would alter 901 # the area above user stack pointer in true asynchronous manner... 902 # 903 # All the above means that if assembler programmer adheres to Unix 904 # register and stack layout, but disregards the "red zone" existense, 905 # it's possible to use following prologue and epilogue to "gear" from 906 # Unix to Win64 ABI in leaf functions with not more than 6 arguments. 907 # 908 # omnipotent_function: 909 # ifdef WIN64 910 # movq %rdi,8(%rsp) 911 # movq %rsi,16(%rsp) 912 # movq %rcx,%rdi ; if 1st argument is actually present 913 # movq %rdx,%rsi ; if 2nd argument is actually ... 914 # movq %r8,%rdx ; if 3rd argument is ... 915 # movq %r9,%rcx ; if 4th argument ... 916 # movq 40(%rsp),%r8 ; if 5th ... 917 # movq 48(%rsp),%r9 ; if 6th ... 918 # endif 919 # ... 920 # ifdef WIN64 921 # movq 8(%rsp),%rdi 922 # movq 16(%rsp),%rsi 923 # endif 924 # ret 925 # 926 ################################################# 928 # Win64 SEH, Structured Exception Handling. 929 # 930 # Unlike on Unix systems(*) lack of Win64 stack unwinding information 931 # has undesired side-effect at run-time: if an exception is raised in 932 # assembler subroutine such as those in question (basically we're 933 # referring to segmentation violations caused by malformed input 934 # parameters), the application is briskly terminated without invoking 935 # any exception handlers, most notably without generating memory dump 936 # or any user notification whatsoever. This poses a problem. It's 937 # possible to address it by registering custom language-specific 938 # handler that would restore processor context to the state at 939 # subroutine entry point and return "exception is not handled, keep 940 # unwinding" code. Writing such handler can be a challenge... But it's 941 # doable, though requires certain coding convention. Consider following 942 # snippet: 943 # 944 # .type function,@function 945 # function: 946 # movq %rsp,%rax # copy rsp to volatile register 947 # pushq %r15 # save non-volatile registers 948 # pushq %rbx 949 # pushq %rbp 950 # movq %rsp,%r11 951 # subq %rdi,%r11 # prepare [variable] stack frame 952 # andq $-64,%r11 953 # movq %rax,0(%r11) # check for exceptions 954 # movq %r11,%rsp # allocate [variable] stack frame 955 # movq %rax,0(%rsp) # save original rsp value 956 # magic_point: 957 # ... 958 # movq 0(%rsp),%rcx # pull original rsp value 959 # movq -24(%rcx),%rbp # restore non-volatile registers 960 # movq -16(%rcx),%rbx 961 # movq -8(%rcx),%r15 962 # movq %rcx,%rsp # restore original rsp 963 # ret 964 # .size function,.-function 965 # 966 # The key is that up to magic_point copy of original rsp value remains 967 # in chosen volatile register and no non-volatile register, except for 968 # rsp, is modified. While past magic_point rsp remains constant till 969 # the very end of the function. In this case custom language-specific 970 # exception handler would look like this: 971 # 972 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, 973 # CONTEXT *context,DISPATCHER_CONTEXT *disp) 974 # { ULONG64 *rsp = (ULONG64 *)context->Rax; 975 # if (context->Rip >= magic_point) 976 # { rsp = ((ULONG64 **)context->Rsp)[0]; 977 # context->Rbp = rsp[-3]; 978 # context->Rbx = rsp[-2]; 979 # context->R15 = rsp[-1]; 980 # } 981 # context->Rsp = (ULONG64)rsp; 982 # context->Rdi = rsp[1]; 983 # context->Rsi = rsp[2]; 984 # 985 # memcpy (disp->ContextRecord,context,sizeof(CONTEXT)); 986 # RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase, 987 # dips->ControlPc,disp->FunctionEntry,disp->ContextRecord, 988 # &disp->HandlerData,&disp->EstablisherFrame,NULL); 989 # return ExceptionContinueSearch; 990 # } 991 # 992 # It's appropriate to implement this handler in assembler, directly in 993 # function's module. In order to do that one has to know members' 994 # offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant 995 # values. Here they are: 996 # 997 # CONTEXT.Rax 120 998 # CONTEXT.Rcx 128 999 # CONTEXT.Rdx 136 1000 # CONTEXT.Rbx 144 1001 # CONTEXT.Rsp 152 1002 # CONTEXT.Rbp 160 1003 # CONTEXT.Rsi 168 1004 # CONTEXT.Rdi 176 1005 # CONTEXT.R8 184 1006 # CONTEXT.R9 192 1007 # CONTEXT.R10 200 1008 # CONTEXT.R11 208 1009 # CONTEXT.R12 216 1010 # CONTEXT.R13 224 1011 # CONTEXT.R14 232 1012 # CONTEXT.R15 240 1013 # CONTEXT.Rip 248 1014 # CONTEXT.Xmm6 512 1015 # sizeof(CONTEXT) 1232 1016 # DISPATCHER_CONTEXT.ControlPc 0 1017 # DISPATCHER_CONTEXT.ImageBase 8 1018 # DISPATCHER_CONTEXT.FunctionEntry 16 1019 # DISPATCHER_CONTEXT.EstablisherFrame 24 1020 # DISPATCHER_CONTEXT.TargetIp 32 1021 # DISPATCHER_CONTEXT.ContextRecord 40 1022 # DISPATCHER_CONTEXT.LanguageHandler 48 1023 # DISPATCHER_CONTEXT.HandlerData 56 1024 # UNW_FLAG_NHANDLER 0 1025 # ExceptionContinueSearch 1 1026 # 1027 # In order to tie the handler to the function one has to compose 1028 # couple of structures: one for .xdata segment and one for .pdata. 1029 # 1030 # UNWIND_INFO structure for .xdata segment would be 1031 # 1032 # function_unwind_info: 1033 # .byte 9,0,0,0 1034 # .rva handler 1035 # 1036 # This structure designates exception handler for a function with 1037 # zero-length prologue, no stack frame or frame register. 1038 # 1039 # To facilitate composing of .pdata structures, auto-generated "gear" 1040 # prologue copies rsp value to rax and denotes next instruction with 1041 # .LSEH_begin_{function_name} label. This essentially defines the SEH 1042 # styling rule mentioned in the beginning. Position of this label is 1043 # chosen in such manner that possible exceptions raised in the "gear" 1044 # prologue would be accounted to caller and unwound from latter's frame. 1045 # End of function is marked with respective .LSEH_end_{function_name} 1046 # label. To summarize, .pdata segment would contain 1047 # 1048 # .rva .LSEH_begin_function 1049 # .rva .LSEH_end_function 1050 # .rva function_unwind_info 1051 # 1052 # Reference to functon_unwind_info from .xdata segment is the anchor. 1053 # In case you wonder why references are 32-bit .rvas and not 64-bit 1054 # .quads. References put into these two segments are required to be 1055 # *relative* to the base address of the current binary module, a.k.a. 1056 # image base. No Win64 module, be it .exe or .dll, can be larger than 1057 # 2GB and thus such relative references can be and are accommodated in 1058 # 32 bits. 1059 # 1060 # Having reviewed the example function code, one can argue that "movq 1061 # %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix 1062 # rax would contain an undefined value. If this "offends" you, use 1063 # another register and refrain from modifying rax till magic_point is 1064 # reached, i.e. as if it was a non-volatile register. If more registers 1065 # are required prior [variable] frame setup is completed, note that 1066 # nobody says that you can have only one "magic point." You can 1067 # "liberate" non-volatile registers by denoting last stack off-load 1068 # instruction and reflecting it in finer grade unwind logic in handler. 1069 # After all, isn't it why it's called *language-specific* handler... 1070 # 1071 # Attentive reader can notice that exceptions would be mishandled in 1072 # auto-generated "gear" epilogue. Well, exception effectively can't 1073 # occur there, because if memory area used by it was subject to 1074 # segmentation violation, then it would be raised upon call to the 1075 # function (and as already mentioned be accounted to caller, which is 1076 # not a problem). If you're still not comfortable, then define tail 1077 # "magic point" just prior ret instruction and have handler treat it... 1078 # 1079 # (*) Note that we're talking about run-time, not debug-time. Lack of 1080 # unwind information makes debugging hard on both Windows and 1081 # Unix. "Unlike" referes to the fact that on Unix signal handler 1082 # will always be invoked, core dumped and appropriate exit code 1083 # returned to parent (for user notification). 1084
