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 61my $flavour = shift; 62my $output = shift; 63if ($flavour =~ /\./) { $output = $flavour; undef $flavour; } 64 65open STDOUT,">$output" || die "can't open $output: $!" 66 if (defined($output)); 67 68my $gas=1; $gas=0 if ($output =~ /\.asm$/); 69my $elf=1; $elf=0 if (!$gas); 70my $win64=0; 71my $prefix=""; 72my $decor=".L"; 73 74my $masmref=8 + 50727*2**-32; # 8.00.50727 shipped with VS2005 75my $masm=0; 76my $PTR=" PTR"; 77 78my $nasmref=2.03; 79my $nasm=0; 80 81if ($flavour eq "mingw64") { $gas=1; $elf=0; $win64=1; 82 $prefix=`echo __USER_LABEL_PREFIX__ | $ENV{CC} -E -P -`; 83 chomp($prefix); 84 } 85elsif ($flavour eq "macosx") { $gas=1; $elf=0; $prefix="_"; $decor="L\$"; } 86elsif ($flavour eq "masm") { $gas=0; $elf=0; $masm=$masmref; $win64=1; $decor="\$L\$"; } 87elsif ($flavour eq "nasm") { $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $decor="\$L\$"; $PTR=""; } 88elsif (!$gas) 89{ if ($ENV{ASM} =~ m/nasm/ && `nasm -v` =~ m/version ([0-9]+)\.([0-9]+)/i) 90 { $nasm = $1 + $2*0.01; $PTR=""; } 91 elsif (`ml64 2>&1` =~ m/Version ([0-9]+)\.([0-9]+)(\.([0-9]+))?/) 92 { $masm = $1 + $2*2**-16 + $4*2**-32; } 93 die "no assembler found on %PATH" if (!($nasm || $masm)); 94 $win64=1; 95 $elf=0; 96 $decor="\$L\$"; 97} 98 99my $current_segment; 100my $current_function; 101my %globals; 102 103{ package opcode; # pick up opcodes 104 sub re { 105 my $self = shift; # single instance in enough... 106 local *line = shift; 107 undef $ret; 108 109 if ($line =~ /^([a-z][a-z0-9]*)/i) { 110 $self->{op} = $1; 111 $ret = $self; 112 $line = substr($line,@+[0]); $line =~ s/^\s+//; 113 114 undef $self->{sz}; 115 if ($self->{op} =~ /^(movz)x?([bw]).*/) { # movz is pain... 116 $self->{op} = $1; 117 $self->{sz} = $2; 118 } elsif ($self->{op} =~ /call|jmp/) { 119 $self->{sz} = ""; 120 } elsif ($self->{op} =~ /^p/ && $' !~ /^(ush|op|insrw)/) { # SSEn 121 $self->{sz} = ""; 122 } elsif ($self->{op} =~ /^v/) { # VEX 123 $self->{sz} = ""; 124 } elsif ($self->{op} =~ /mov[dq]/ && $line =~ /%xmm/) { 125 $self->{sz} = ""; 126 } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) { 127 $self->{op} = $1; 128 $self->{sz} = $2; 129 } 130 } 131 $ret; 132 } 133 sub size { 134 my $self = shift; 135 my $sz = shift; 136 $self->{sz} = $sz if (defined($sz) && !defined($self->{sz})); 137 $self->{sz}; 138 } 139 sub out { 140 my $self = shift; 141 if ($gas) { 142 if ($self->{op} eq "movz") { # movz is pain... 143 sprintf "%s%s%s",$self->{op},$self->{sz},shift; 144 } elsif ($self->{op} =~ /^set/) { 145 "$self->{op}"; 146 } elsif ($self->{op} eq "ret") { 147 my $epilogue = ""; 148 if ($win64 && $current_function->{abi} eq "svr4") { 149 $epilogue = "movq 8(%rsp),%rdi\n\t" . 150 "movq 16(%rsp),%rsi\n\t"; 151 } 152 $epilogue . ".byte 0xf3,0xc3"; 153 } elsif ($self->{op} eq "call" && !$elf && $current_segment eq ".init") { 154 ".p2align\t3\n\t.quad"; 155 } else { 156 "$self->{op}$self->{sz}"; 157 } 158 } else { 159 $self->{op} =~ s/^movz/movzx/; 160 if ($self->{op} eq "ret") { 161 $self->{op} = ""; 162 if ($win64 && $current_function->{abi} eq "svr4") { 163 $self->{op} = "mov rdi,QWORD${PTR}[8+rsp]\t;WIN64 epilogue\n\t". 164 "mov rsi,QWORD${PTR}[16+rsp]\n\t"; 165 } 166 $self->{op} .= "DB\t0F3h,0C3h\t\t;repret"; 167 } elsif ($self->{op} =~ /^(pop|push)f/) { 168 $self->{op} .= $self->{sz}; 169 } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") { 170 $self->{op} = "\tDQ"; 171 } 172 $self->{op}; 173 } 174 } 175 sub mnemonic { 176 my $self=shift; 177 my $op=shift; 178 $self->{op}=$op if (defined($op)); 179 $self->{op}; 180 } 181} 182{ package const; # pick up constants, which start with $ 183 sub re { 184 my $self = shift; # single instance in enough... 185 local *line = shift; 186 undef $ret; 187 188 if ($line =~ /^\$([^,]+)/) { 189 $self->{value} = $1; 190 $ret = $self; 191 $line = substr($line,@+[0]); $line =~ s/^\s+//; 192 } 193 $ret; 194 } 195 sub out { 196 my $self = shift; 197 198 $self->{value} =~ s/\b(0b[0-1]+)/oct($1)/eig; 199 if ($gas) { 200 # Solaris /usr/ccs/bin/as can't handle multiplications 201 # in $self->{value} 202 my $value = $self->{value}; 203 $value =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi; 204 if ($value =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg) { 205 $self->{value} = $value; 206 } 207 sprintf "\$%s",$self->{value}; 208 } else { 209 $self->{value} =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm); 210 sprintf "%s",$self->{value}; 211 } 212 } 213} 214{ package ea; # pick up effective addresses: expr(%reg,%reg,scale) 215 sub re { 216 my $self = shift; # single instance in enough... 217 local *line = shift; 218 undef $ret; 219 220 # optional * ---vvv--- appears in indirect jmp/call 221 if ($line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)/) { 222 $self->{asterisk} = $1; 223 $self->{label} = $2; 224 ($self->{base},$self->{index},$self->{scale})=split(/,/,$3); 225 $self->{scale} = 1 if (!defined($self->{scale})); 226 $ret = $self; 227 $line = substr($line,@+[0]); $line =~ s/^\s+//; 228 229 if ($win64 && $self->{label} =~ s/\@GOTPCREL//) { 230 die if (opcode->mnemonic() ne "mov"); 231 opcode->mnemonic("lea"); 232 } 233 $self->{base} =~ s/^%//; 234 $self->{index} =~ s/^%// if (defined($self->{index})); 235 } 236 $ret; 237 } 238 sub size {} 239 sub out { 240 my $self = shift; 241 my $sz = shift; 242 243 $self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 244 $self->{label} =~ s/\.L/$decor/g; 245 246 # Silently convert all EAs to 64-bit. This is required for 247 # elder GNU assembler and results in more compact code, 248 # *but* most importantly AES module depends on this feature! 249 $self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; 250 $self->{base} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; 251 252 # Solaris /usr/ccs/bin/as can't handle multiplications 253 # in $self->{label}, new gas requires sign extension... 254 use integer; 255 $self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi; 256 $self->{label} =~ s/\b([0-9]+\s*[\*\/\%]\s*[0-9]+)\b/eval($1)/eg; 257 $self->{label} =~ s/\b([0-9]+)\b/$1<<32>>32/eg; 258 259 if (!$self->{label} && $self->{index} && $self->{scale}==1 && 260 $self->{base} =~ /(rbp|r13)/) { 261 $self->{base} = $self->{index}; $self->{index} = $1; 262 } 263 264 if ($gas) { 265 $self->{label} =~ s/^___imp_/__imp__/ if ($flavour eq "mingw64"); 266 267 if (defined($self->{index})) { 268 sprintf "%s%s(%s,%%%s,%d)",$self->{asterisk}, 269 $self->{label}, 270 $self->{base}?"%$self->{base}":"", 271 $self->{index},$self->{scale}; 272 } else { 273 sprintf "%s%s(%%%s)", $self->{asterisk},$self->{label},$self->{base}; 274 } 275 } else { 276 %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR", 277 l=>"DWORD$PTR", d=>"DWORD$PTR", 278 q=>"QWORD$PTR", o=>"OWORD$PTR", 279 x=>"XMMWORD$PTR", y=>"YMMWORD$PTR", z=>"ZMMWORD$PTR" ); 280 281 $self->{label} =~ s/\./\$/g; 282 $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig; 283 $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/); 284 285 ($self->{asterisk}) && ($sz="q") || 286 (opcode->mnemonic() =~ /^v?mov([qd])$/) && ($sz=$1) || 287 (opcode->mnemonic() =~ /^v?pinsr([qdwb])$/) && ($sz=$1) || 288 (opcode->mnemonic() =~ /^vpbroadcast([qdwb])$/) && ($sz=$1) || 289 (opcode->mnemonic() =~ /^vinsert[fi]128$/) && ($sz="x"); 290 291 if (defined($self->{index})) { 292 sprintf "%s[%s%s*%d%s]",$szmap{$sz}, 293 $self->{label}?"$self->{label}+":"", 294 $self->{index},$self->{scale}, 295 $self->{base}?"+$self->{base}":""; 296 } elsif ($self->{base} eq "rip") { 297 sprintf "%s[%s]",$szmap{$sz},$self->{label}; 298 } else { 299 sprintf "%s[%s%s]",$szmap{$sz}, 300 $self->{label}?"$self->{label}+":"", 301 $self->{base}; 302 } 303 } 304 } 305} 306{ package register; # pick up registers, which start with %. 307 sub re { 308 my $class = shift; # muliple instances... 309 my $self = {}; 310 local *line = shift; 311 undef $ret; 312 313 # optional * ---vvv--- appears in indirect jmp/call 314 if ($line =~ /^(\*?)%(\w+)/) { 315 bless $self,$class; 316 $self->{asterisk} = $1; 317 $self->{value} = $2; 318 $ret = $self; 319 $line = substr($line,@+[0]); $line =~ s/^\s+//; 320 } 321 $ret; 322 } 323 sub size { 324 my $self = shift; 325 undef $ret; 326 327 if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; } 328 elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; } 329 elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; } 330 elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; } 331 elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; } 332 elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; } 333 elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; } 334 elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; } 335 336 $ret; 337 } 338 sub out { 339 my $self = shift; 340 if ($gas) { sprintf "%s%%%s",$self->{asterisk},$self->{value}; } 341 else { $self->{value}; } 342 } 343} 344{ package label; # pick up labels, which end with : 345 sub re { 346 my $self = shift; # single instance is enough... 347 local *line = shift; 348 undef $ret; 349 350 if ($line =~ /(^[\.\w]+)\:/) { 351 $self->{value} = $1; 352 $ret = $self; 353 $line = substr($line,@+[0]); $line =~ s/^\s+//; 354 355 $self->{value} =~ s/^\.L/$decor/; 356 } 357 $ret; 358 } 359 sub out { 360 my $self = shift; 361 362 if ($gas) { 363 my $func = ($globals{$self->{value}} or $self->{value}) . ":"; 364 if ($win64 && 365 $current_function->{name} eq $self->{value} && 366 $current_function->{abi} eq "svr4") { 367 $func .= "\n"; 368 $func .= " movq %rdi,8(%rsp)\n"; 369 $func .= " movq %rsi,16(%rsp)\n"; 370 $func .= " movq %rsp,%rax\n"; 371 $func .= "${decor}SEH_begin_$current_function->{name}:\n"; 372 my $narg = $current_function->{narg}; 373 $narg=6 if (!defined($narg)); 374 $func .= " movq %rcx,%rdi\n" if ($narg>0); 375 $func .= " movq %rdx,%rsi\n" if ($narg>1); 376 $func .= " movq %r8,%rdx\n" if ($narg>2); 377 $func .= " movq %r9,%rcx\n" if ($narg>3); 378 $func .= " movq 40(%rsp),%r8\n" if ($narg>4); 379 $func .= " movq 48(%rsp),%r9\n" if ($narg>5); 380 } 381 $func; 382 } elsif ($self->{value} ne "$current_function->{name}") { 383 $self->{value} .= ":" if ($masm && $ret!~m/^\$/); 384 $self->{value} . ":"; 385 } elsif ($win64 && $current_function->{abi} eq "svr4") { 386 my $func = "$current_function->{name}" . 387 ($nasm ? ":" : "\tPROC $current_function->{scope}") . 388 "\n"; 389 $func .= " mov QWORD${PTR}[8+rsp],rdi\t;WIN64 prologue\n"; 390 $func .= " mov QWORD${PTR}[16+rsp],rsi\n"; 391 $func .= " mov rax,rsp\n"; 392 $func .= "${decor}SEH_begin_$current_function->{name}:"; 393 $func .= ":" if ($masm); 394 $func .= "\n"; 395 my $narg = $current_function->{narg}; 396 $narg=6 if (!defined($narg)); 397 $func .= " mov rdi,rcx\n" if ($narg>0); 398 $func .= " mov rsi,rdx\n" if ($narg>1); 399 $func .= " mov rdx,r8\n" if ($narg>2); 400 $func .= " mov rcx,r9\n" if ($narg>3); 401 $func .= " mov r8,QWORD${PTR}[40+rsp]\n" if ($narg>4); 402 $func .= " mov r9,QWORD${PTR}[48+rsp]\n" if ($narg>5); 403 $func .= "\n"; 404 } else { 405 "$current_function->{name}". 406 ($nasm ? ":" : "\tPROC $current_function->{scope}"); 407 } 408 } 409} 410{ package expr; # pick up expressioins 411 sub re { 412 my $self = shift; # single instance is enough... 413 local *line = shift; 414 undef $ret; 415 416 if ($line =~ /(^[^,]+)/) { 417 $self->{value} = $1; 418 $ret = $self; 419 $line = substr($line,@+[0]); $line =~ s/^\s+//; 420 421 $self->{value} =~ s/\@PLT// if (!$elf); 422 $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 423 $self->{value} =~ s/\.L/$decor/g; 424 } 425 $ret; 426 } 427 sub out { 428 my $self = shift; 429 if ($nasm && opcode->mnemonic()=~m/^j(?![re]cxz)/) { 430 "NEAR ".$self->{value}; 431 } else { 432 $self->{value}; 433 } 434 } 435} 436{ package directive; # pick up directives, which start with . 437 sub re { 438 my $self = shift; # single instance is enough... 439 local *line = shift; 440 undef $ret; 441 my $dir; 442 my %opcode = # lea 2f-1f(%rip),%dst; 1: nop; 2: 443 ( "%rax"=>0x01058d48, "%rcx"=>0x010d8d48, 444 "%rdx"=>0x01158d48, "%rbx"=>0x011d8d48, 445 "%rsp"=>0x01258d48, "%rbp"=>0x012d8d48, 446 "%rsi"=>0x01358d48, "%rdi"=>0x013d8d48, 447 "%r8" =>0x01058d4c, "%r9" =>0x010d8d4c, 448 "%r10"=>0x01158d4c, "%r11"=>0x011d8d4c, 449 "%r12"=>0x01258d4c, "%r13"=>0x012d8d4c, 450 "%r14"=>0x01358d4c, "%r15"=>0x013d8d4c ); 451 452 if ($line =~ /^\s*(\.\w+)/) { 453 $dir = $1; 454 $ret = $self; 455 undef $self->{value}; 456 $line = substr($line,@+[0]); $line =~ s/^\s+//; 457 458 SWITCH: for ($dir) { 459 /\.picmeup/ && do { if ($line =~ /(%r[\w]+)/i) { 460 $dir="\t.long"; 461 $line=sprintf "0x%x,0x90000000",$opcode{$1}; 462 } 463 last; 464 }; 465 /\.global|\.globl|\.extern/ 466 && do { $globals{$line} = $prefix . $line; 467 $line = $globals{$line} if ($prefix); 468 last; 469 }; 470 /\.type/ && do { ($sym,$type,$narg) = split(',',$line); 471 if ($type eq "\@function") { 472 undef $current_function; 473 $current_function->{name} = $sym; 474 $current_function->{abi} = "svr4"; 475 $current_function->{narg} = $narg; 476 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 477 } elsif ($type eq "\@abi-omnipotent") { 478 undef $current_function; 479 $current_function->{name} = $sym; 480 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 481 } 482 $line =~ s/\@abi\-omnipotent/\@function/; 483 $line =~ s/\@function.*/\@function/; 484 last; 485 }; 486 /\.asciz/ && do { if ($line =~ /^"(.*)"$/) { 487 $dir = ".byte"; 488 $line = join(",",unpack("C*",$1),0); 489 } 490 last; 491 }; 492 /\.rva|\.long|\.quad/ 493 && do { $line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 494 $line =~ s/\.L/$decor/g; 495 last; 496 }; 497 } 498 499 if ($gas) { 500 $self->{value} = $dir . "\t" . $line; 501 502 if ($dir =~ /\.extern/) { 503 $self->{value} = ""; # swallow extern 504 } elsif (!$elf && $dir =~ /\.type/) { 505 $self->{value} = ""; 506 $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" . 507 (defined($globals{$1})?".scl 2;":".scl 3;") . 508 "\t.type 32;\t.endef" 509 if ($win64 && $line =~ /([^,]+),\@function/); 510 } elsif (!$elf && $dir =~ /\.size/) { 511 $self->{value} = ""; 512 if (defined($current_function)) { 513 $self->{value} .= "${decor}SEH_end_$current_function->{name}:" 514 if ($win64 && $current_function->{abi} eq "svr4"); 515 undef $current_function; 516 } 517 } elsif (!$elf && $dir =~ /\.align/) { 518 $self->{value} = ".p2align\t" . (log($line)/log(2)); 519 } elsif ($dir eq ".section") { 520 $current_segment=$line; 521 if (!$elf && $current_segment eq ".init") { 522 if ($flavour eq "macosx") { $self->{value} = ".mod_init_func"; } 523 elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.ctors"; } 524 } 525 } elsif ($dir =~ /\.(text|data)/) { 526 $current_segment=".$1"; 527 } elsif ($dir =~ /\.hidden/) { 528 if ($flavour eq "macosx") { $self->{value} = ".private_extern\t$prefix$line"; } 529 elsif ($flavour eq "mingw64") { $self->{value} = ""; } 530 } elsif ($dir =~ /\.comm/) { 531 $self->{value} = "$dir\t$prefix$line"; 532 $self->{value} =~ s|,([0-9]+),([0-9]+)$|",$1,".log($2)/log(2)|e if ($flavour eq "macosx"); 533 } 534 $line = ""; 535 return $self; 536 } 537 538 # non-gas case or nasm/masm 539 SWITCH: for ($dir) { 540 /\.text/ && do { my $v=undef; 541 if ($nasm) { 542 $v="section .text code align=64\n"; 543 } else { 544 $v="$current_segment\tENDS\n" if ($current_segment); 545 $current_segment = ".text\$"; 546 $v.="$current_segment\tSEGMENT "; 547 $v.=$masm>=$masmref ? "ALIGN(256)" : "PAGE"; 548 $v.=" 'CODE'"; 549 } 550 $self->{value} = $v; 551 last; 552 }; 553 /\.data/ && do { my $v=undef; 554 if ($nasm) { 555 $v="section .data data align=8\n"; 556 } else { 557 $v="$current_segment\tENDS\n" if ($current_segment); 558 $current_segment = "_DATA"; 559 $v.="$current_segment\tSEGMENT"; 560 } 561 $self->{value} = $v; 562 last; 563 }; 564 /\.section/ && do { my $v=undef; 565 $line =~ s/([^,]*).*/$1/; 566 $line = ".CRT\$XCU" if ($line eq ".init"); 567 if ($nasm) { 568 $v="section $line"; 569 if ($line=~/\.([px])data/) { 570 $v.=" rdata align="; 571 $v.=$1 eq "p"? 4 : 8; 572 } elsif ($line=~/\.CRT\$/i) { 573 $v.=" rdata align=8"; 574 } 575 } else { 576 $v="$current_segment\tENDS\n" if ($current_segment); 577 $v.="$line\tSEGMENT"; 578 if ($line=~/\.([px])data/) { 579 $v.=" READONLY"; 580 $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref); 581 } elsif ($line=~/\.CRT\$/i) { 582 $v.=" READONLY "; 583 $v.=$masm>=$masmref ? "ALIGN(8)" : "DWORD"; 584 } 585 } 586 $current_segment = $line; 587 $self->{value} = $v; 588 last; 589 }; 590 /\.extern/ && do { $self->{value} = "EXTERN\t".$line; 591 $self->{value} .= ":NEAR" if ($masm); 592 last; 593 }; 594 /\.globl|.global/ 595 && do { $self->{value} = $masm?"PUBLIC":"global"; 596 $self->{value} .= "\t".$line; 597 last; 598 }; 599 /\.size/ && do { if (defined($current_function)) { 600 undef $self->{value}; 601 if ($current_function->{abi} eq "svr4") { 602 $self->{value}="${decor}SEH_end_$current_function->{name}:"; 603 $self->{value}.=":\n" if($masm); 604 } 605 $self->{value}.="$current_function->{name}\tENDP" if($masm && $current_function->{name}); 606 undef $current_function; 607 } 608 last; 609 }; 610 /\.align/ && do { $self->{value} = "ALIGN\t".$line; last; }; 611 /\.(value|long|rva|quad)/ 612 && do { my $sz = substr($1,0,1); 613 my @arr = split(/,\s*/,$line); 614 my $last = pop(@arr); 615 my $conv = sub { my $var=shift; 616 $var=~s/^(0b[0-1]+)/oct($1)/eig; 617 $var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm); 618 if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva")) 619 { $var=~s/([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; } 620 $var; 621 }; 622 623 $sz =~ tr/bvlrq/BWDDQ/; 624 $self->{value} = "\tD$sz\t"; 625 for (@arr) { $self->{value} .= &$conv($_).","; } 626 $self->{value} .= &$conv($last); 627 last; 628 }; 629 /\.byte/ && do { my @str=split(/,\s*/,$line); 630 map(s/(0b[0-1]+)/oct($1)/eig,@str); 631 map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm); 632 while ($#str>15) { 633 $self->{value}.="DB\t" 634 .join(",",@str[0..15])."\n"; 635 foreach (0..15) { shift @str; } 636 } 637 $self->{value}.="DB\t" 638 .join(",",@str) if (@str); 639 last; 640 }; 641 /\.comm/ && do { my @str=split(/,\s*/,$line); 642 my $v=undef; 643 if ($nasm) { 644 $v.="common $prefix@str[0] @str[1]"; 645 } else { 646 $v="$current_segment\tENDS\n" if ($current_segment); 647 $current_segment = "_DATA"; 648 $v.="$current_segment\tSEGMENT\n"; 649 $v.="COMM @str[0]:DWORD:".@str[1]/4; 650 } 651 $self->{value} = $v; 652 last; 653 }; 654 } 655 $line = ""; 656 } 657 658 $ret; 659 } 660 sub out { 661 my $self = shift; 662 $self->{value}; 663 } 664} 665 666sub rex { 667 local *opcode=shift; 668 my ($dst,$src,$rex)=@_; 669 670 $rex|=0x04 if($dst>=8); 671 $rex|=0x01 if($src>=8); 672 push @opcode,($rex|0x40) if ($rex); 673} 674 675# older gas and ml64 don't handle SSE>2 instructions 676my %regrm = ( "%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3, 677 "%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7 ); 678 679my $movq = sub { # elderly gas can't handle inter-register movq 680 my $arg = shift; 681 my @opcode=(0x66); 682 if ($arg =~ /%xmm([0-9]+),\s*%r(\w+)/) { 683 my ($src,$dst)=($1,$2); 684 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 685 rex(\@opcode,$src,$dst,0x8); 686 push @opcode,0x0f,0x7e; 687 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 688 @opcode; 689 } elsif ($arg =~ /%r(\w+),\s*%xmm([0-9]+)/) { 690 my ($src,$dst)=($2,$1); 691 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 692 rex(\@opcode,$src,$dst,0x8); 693 push @opcode,0x0f,0x6e; 694 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 695 @opcode; 696 } else { 697 (); 698 } 699}; 700 701my $pextrd = sub { 702 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*(%\w+)/) { 703 my @opcode=(0x66); 704 $imm=$1; 705 $src=$2; 706 $dst=$3; 707 if ($dst =~ /%r([0-9]+)d/) { $dst = $1; } 708 elsif ($dst =~ /%e/) { $dst = $regrm{$dst}; } 709 rex(\@opcode,$src,$dst); 710 push @opcode,0x0f,0x3a,0x16; 711 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 712 push @opcode,$imm; 713 @opcode; 714 } else { 715 (); 716 } 717}; 718 719my $pinsrd = sub { 720 if (shift =~ /\$([0-9]+),\s*(%\w+),\s*%xmm([0-9]+)/) { 721 my @opcode=(0x66); 722 $imm=$1; 723 $src=$2; 724 $dst=$3; 725 if ($src =~ /%r([0-9]+)/) { $src = $1; } 726 elsif ($src =~ /%e/) { $src = $regrm{$src}; } 727 rex(\@opcode,$dst,$src); 728 push @opcode,0x0f,0x3a,0x22; 729 push @opcode,0xc0|(($dst&7)<<3)|($src&7); # ModR/M 730 push @opcode,$imm; 731 @opcode; 732 } else { 733 (); 734 } 735}; 736 737my $pshufb = sub { 738 if (shift =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) { 739 my @opcode=(0x66); 740 rex(\@opcode,$2,$1); 741 push @opcode,0x0f,0x38,0x00; 742 push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M 743 @opcode; 744 } else { 745 (); 746 } 747}; 748 749my $palignr = sub { 750 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 751 my @opcode=(0x66); 752 rex(\@opcode,$3,$2); 753 push @opcode,0x0f,0x3a,0x0f; 754 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 755 push @opcode,$1; 756 @opcode; 757 } else { 758 (); 759 } 760}; 761 762my $pclmulqdq = sub { 763 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 764 my @opcode=(0x66); 765 rex(\@opcode,$3,$2); 766 push @opcode,0x0f,0x3a,0x44; 767 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 768 my $c=$1; 769 push @opcode,$c=~/^0/?oct($c):$c; 770 @opcode; 771 } else { 772 (); 773 } 774}; 775 776my $rdrand = sub { 777 if (shift =~ /%[er](\w+)/) { 778 my @opcode=(); 779 my $dst=$1; 780 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 781 rex(\@opcode,0,$1,8); 782 push @opcode,0x0f,0xc7,0xf0|($dst&7); 783 @opcode; 784 } else { 785 (); 786 } 787}; 788 789my $rdseed = sub { 790 if (shift =~ /%[er](\w+)/) { 791 my @opcode=(); 792 my $dst=$1; 793 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 794 rex(\@opcode,0,$1,8); 795 push @opcode,0x0f,0xc7,0xf8|($dst&7); 796 @opcode; 797 } else { 798 (); 799 } 800}; 801 802sub rxb { 803 local *opcode=shift; 804 my ($dst,$src1,$src2,$rxb)=@_; 805 806 $rxb|=0x7<<5; 807 $rxb&=~(0x04<<5) if($dst>=8); 808 $rxb&=~(0x01<<5) if($src1>=8); 809 $rxb&=~(0x02<<5) if($src2>=8); 810 push @opcode,$rxb; 811} 812 813my $vprotd = sub { 814 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 815 my @opcode=(0x8f); 816 rxb(\@opcode,$3,$2,-1,0x08); 817 push @opcode,0x78,0xc2; 818 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 819 my $c=$1; 820 push @opcode,$c=~/^0/?oct($c):$c; 821 @opcode; 822 } else { 823 (); 824 } 825}; 826 827my $vprotq = sub { 828 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 829 my @opcode=(0x8f); 830 rxb(\@opcode,$3,$2,-1,0x08); 831 push @opcode,0x78,0xc3; 832 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 833 my $c=$1; 834 push @opcode,$c=~/^0/?oct($c):$c; 835 @opcode; 836 } else { 837 (); 838 } 839}; 840 841if ($nasm) { 842 print <<___; 843default rel 844%define XMMWORD 845%define YMMWORD 846%define ZMMWORD 847___ 848} elsif ($masm) { 849 print <<___; 850OPTION DOTNAME 851___ 852} 853while($line=<>) { 854 855 chomp($line); 856 857 $line =~ s|[#!].*$||; # get rid of asm-style comments... 858 $line =~ s|/\*.*\*/||; # ... and C-style comments... 859 $line =~ s|^\s+||; # ... and skip white spaces in beginning 860 $line =~ s|\s+$||; # ... and at the end 861 862 undef $label; 863 undef $opcode; 864 undef @args; 865 866 if ($label=label->re(\$line)) { print $label->out(); } 867 868 if (directive->re(\$line)) { 869 printf "%s",directive->out(); 870 } elsif ($opcode=opcode->re(\$line)) { 871 my $asm = eval("\$".$opcode->mnemonic()); 872 undef @bytes; 873 874 if ((ref($asm) eq 'CODE') && scalar(@bytes=&$asm($line))) { 875 print $gas?".byte\t":"DB\t",join(',',@bytes),"\n"; 876 next; 877 } 878 879 ARGUMENT: while (1) { 880 my $arg; 881 882 if ($arg=register->re(\$line)) { opcode->size($arg->size()); } 883 elsif ($arg=const->re(\$line)) { } 884 elsif ($arg=ea->re(\$line)) { } 885 elsif ($arg=expr->re(\$line)) { } 886 else { last ARGUMENT; } 887 888 push @args,$arg; 889 890 last ARGUMENT if ($line !~ /^,/); 891 892 $line =~ s/^,\s*//; 893 } # ARGUMENT: 894 895 if ($#args>=0) { 896 my $insn; 897 my $sz=opcode->size(); 898 899 if ($gas) { 900 $insn = $opcode->out($#args>=1?$args[$#args]->size():$sz); 901 @args = map($_->out($sz),@args); 902 printf "\t%s\t%s",$insn,join(",",@args); 903 } else { 904 $insn = $opcode->out(); 905 foreach (@args) { 906 my $arg = $_->out(); 907 # $insn.=$sz compensates for movq, pinsrw, ... 908 if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; } 909 if ($arg =~ /^ymm[0-9]+$/) { $insn.=$sz; $sz="y" if(!$sz); last; } 910 if ($arg =~ /^zmm[0-9]+$/) { $insn.=$sz; $sz="z" if(!$sz); last; } 911 if ($arg =~ /^mm[0-9]+$/) { $insn.=$sz; $sz="q" if(!$sz); last; } 912 } 913 @args = reverse(@args); 914 undef $sz if ($nasm && $opcode->mnemonic() eq "lea"); 915 printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args)); 916 } 917 } else { 918 printf "\t%s",$opcode->out(); 919 } 920 } 921 922 print $line,"\n"; 923} 924 925print "\n$current_segment\tENDS\n" if ($current_segment && $masm); 926print "END\n" if ($masm); 927 928close STDOUT; 929 930 ################################################# 931# Cross-reference x86_64 ABI "card" 932# 933# Unix Win64 934# %rax * * 935# %rbx - - 936# %rcx #4 #1 937# %rdx #3 #2 938# %rsi #2 - 939# %rdi #1 - 940# %rbp - - 941# %rsp - - 942# %r8 #5 #3 943# %r9 #6 #4 944# %r10 * * 945# %r11 * * 946# %r12 - - 947# %r13 - - 948# %r14 - - 949# %r15 - - 950# 951# (*) volatile register 952# (-) preserved by callee 953# (#) Nth argument, volatile 954# 955# In Unix terms top of stack is argument transfer area for arguments 956# which could not be accomodated in registers. Or in other words 7th 957# [integer] argument resides at 8(%rsp) upon function entry point. 958# 128 bytes above %rsp constitute a "red zone" which is not touched 959# by signal handlers and can be used as temporal storage without 960# allocating a frame. 961# 962# In Win64 terms N*8 bytes on top of stack is argument transfer area, 963# which belongs to/can be overwritten by callee. N is the number of 964# arguments passed to callee, *but* not less than 4! This means that 965# upon function entry point 5th argument resides at 40(%rsp), as well 966# as that 32 bytes from 8(%rsp) can always be used as temporal 967# storage [without allocating a frame]. One can actually argue that 968# one can assume a "red zone" above stack pointer under Win64 as well. 969# Point is that at apparently no occasion Windows kernel would alter 970# the area above user stack pointer in true asynchronous manner... 971# 972# All the above means that if assembler programmer adheres to Unix 973# register and stack layout, but disregards the "red zone" existense, 974# it's possible to use following prologue and epilogue to "gear" from 975# Unix to Win64 ABI in leaf functions with not more than 6 arguments. 976# 977# omnipotent_function: 978# ifdef WIN64 979# movq %rdi,8(%rsp) 980# movq %rsi,16(%rsp) 981# movq %rcx,%rdi ; if 1st argument is actually present 982# movq %rdx,%rsi ; if 2nd argument is actually ... 983# movq %r8,%rdx ; if 3rd argument is ... 984# movq %r9,%rcx ; if 4th argument ... 985# movq 40(%rsp),%r8 ; if 5th ... 986# movq 48(%rsp),%r9 ; if 6th ... 987# endif 988# ... 989# ifdef WIN64 990# movq 8(%rsp),%rdi 991# movq 16(%rsp),%rsi 992# endif 993# ret 994# 995 ################################################# 996# Win64 SEH, Structured Exception Handling. 997# 998# Unlike on Unix systems(*) lack of Win64 stack unwinding information 999# has undesired side-effect at run-time: if an exception is raised in 1000# assembler subroutine such as those in question (basically we're 1001# referring to segmentation violations caused by malformed input 1002# parameters), the application is briskly terminated without invoking 1003# any exception handlers, most notably without generating memory dump 1004# or any user notification whatsoever. This poses a problem. It's 1005# possible to address it by registering custom language-specific 1006# handler that would restore processor context to the state at 1007# subroutine entry point and return "exception is not handled, keep 1008# unwinding" code. Writing such handler can be a challenge... But it's 1009# doable, though requires certain coding convention. Consider following 1010# snippet: 1011# 1012# .type function,@function 1013# function: 1014# movq %rsp,%rax # copy rsp to volatile register 1015# pushq %r15 # save non-volatile registers 1016# pushq %rbx 1017# pushq %rbp 1018# movq %rsp,%r11 1019# subq %rdi,%r11 # prepare [variable] stack frame 1020# andq $-64,%r11 1021# movq %rax,0(%r11) # check for exceptions 1022# movq %r11,%rsp # allocate [variable] stack frame 1023# movq %rax,0(%rsp) # save original rsp value 1024# magic_point: 1025# ... 1026# movq 0(%rsp),%rcx # pull original rsp value 1027# movq -24(%rcx),%rbp # restore non-volatile registers 1028# movq -16(%rcx),%rbx 1029# movq -8(%rcx),%r15 1030# movq %rcx,%rsp # restore original rsp 1031# ret 1032# .size function,.-function 1033# 1034# The key is that up to magic_point copy of original rsp value remains 1035# in chosen volatile register and no non-volatile register, except for 1036# rsp, is modified. While past magic_point rsp remains constant till 1037# the very end of the function. In this case custom language-specific 1038# exception handler would look like this: 1039# 1040# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, 1041# CONTEXT *context,DISPATCHER_CONTEXT *disp) 1042# { ULONG64 *rsp = (ULONG64 *)context->Rax; 1043# if (context->Rip >= magic_point) 1044# { rsp = ((ULONG64 **)context->Rsp)[0]; 1045# context->Rbp = rsp[-3]; 1046# context->Rbx = rsp[-2]; 1047# context->R15 = rsp[-1]; 1048# } 1049# context->Rsp = (ULONG64)rsp; 1050# context->Rdi = rsp[1]; 1051# context->Rsi = rsp[2]; 1052# 1053# memcpy (disp->ContextRecord,context,sizeof(CONTEXT)); 1054# RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase, 1055# dips->ControlPc,disp->FunctionEntry,disp->ContextRecord, 1056# &disp->HandlerData,&disp->EstablisherFrame,NULL); 1057# return ExceptionContinueSearch; 1058# } 1059# 1060# It's appropriate to implement this handler in assembler, directly in 1061# function's module. In order to do that one has to know members' 1062# offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant 1063# values. Here they are: 1064# 1065# CONTEXT.Rax 120 1066# CONTEXT.Rcx 128 1067# CONTEXT.Rdx 136 1068# CONTEXT.Rbx 144 1069# CONTEXT.Rsp 152 1070# CONTEXT.Rbp 160 1071# CONTEXT.Rsi 168 1072# CONTEXT.Rdi 176 1073# CONTEXT.R8 184 1074# CONTEXT.R9 192 1075# CONTEXT.R10 200 1076# CONTEXT.R11 208 1077# CONTEXT.R12 216 1078# CONTEXT.R13 224 1079# CONTEXT.R14 232 1080# CONTEXT.R15 240 1081# CONTEXT.Rip 248 1082# CONTEXT.Xmm6 512 1083# sizeof(CONTEXT) 1232 1084# DISPATCHER_CONTEXT.ControlPc 0 1085# DISPATCHER_CONTEXT.ImageBase 8 1086# DISPATCHER_CONTEXT.FunctionEntry 16 1087# DISPATCHER_CONTEXT.EstablisherFrame 24 1088# DISPATCHER_CONTEXT.TargetIp 32 1089# DISPATCHER_CONTEXT.ContextRecord 40 1090# DISPATCHER_CONTEXT.LanguageHandler 48 1091# DISPATCHER_CONTEXT.HandlerData 56 1092# UNW_FLAG_NHANDLER 0 1093# ExceptionContinueSearch 1 1094# 1095# In order to tie the handler to the function one has to compose 1096# couple of structures: one for .xdata segment and one for .pdata. 1097# 1098# UNWIND_INFO structure for .xdata segment would be 1099# 1100# function_unwind_info: 1101# .byte 9,0,0,0 1102# .rva handler 1103# 1104# This structure designates exception handler for a function with 1105# zero-length prologue, no stack frame or frame register. 1106# 1107# To facilitate composing of .pdata structures, auto-generated "gear" 1108# prologue copies rsp value to rax and denotes next instruction with 1109# .LSEH_begin_{function_name} label. This essentially defines the SEH 1110# styling rule mentioned in the beginning. Position of this label is 1111# chosen in such manner that possible exceptions raised in the "gear" 1112# prologue would be accounted to caller and unwound from latter's frame. 1113# End of function is marked with respective .LSEH_end_{function_name} 1114# label. To summarize, .pdata segment would contain 1115# 1116# .rva .LSEH_begin_function 1117# .rva .LSEH_end_function 1118# .rva function_unwind_info 1119# 1120# Reference to functon_unwind_info from .xdata segment is the anchor. 1121# In case you wonder why references are 32-bit .rvas and not 64-bit 1122# .quads. References put into these two segments are required to be 1123# *relative* to the base address of the current binary module, a.k.a. 1124# image base. No Win64 module, be it .exe or .dll, can be larger than 1125# 2GB and thus such relative references can be and are accommodated in 1126# 32 bits. 1127# 1128# Having reviewed the example function code, one can argue that "movq 1129# %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix 1130# rax would contain an undefined value. If this "offends" you, use 1131# another register and refrain from modifying rax till magic_point is 1132# reached, i.e. as if it was a non-volatile register. If more registers 1133# are required prior [variable] frame setup is completed, note that 1134# nobody says that you can have only one "magic point." You can 1135# "liberate" non-volatile registers by denoting last stack off-load 1136# instruction and reflecting it in finer grade unwind logic in handler. 1137# After all, isn't it why it's called *language-specific* handler... 1138# 1139# Attentive reader can notice that exceptions would be mishandled in 1140# auto-generated "gear" epilogue. Well, exception effectively can't 1141# occur there, because if memory area used by it was subject to 1142# segmentation violation, then it would be raised upon call to the 1143# function (and as already mentioned be accounted to caller, which is 1144# not a problem). If you're still not comfortable, then define tail 1145# "magic point" just prior ret instruction and have handler treat it... 1146# 1147# (*) Note that we're talking about run-time, not debug-time. Lack of 1148# unwind information makes debugging hard on both Windows and 1149# Unix. "Unlike" referes to the fact that on Unix signal handler 1150# will always be invoked, core dumped and appropriate exit code 1151# returned to parent (for user notification). 1152