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 # TODO(davidben): Before supporting the 83 # mingw64 perlasm flavour, do away with this 84 # environment variable check. 85 die "mingw64 not supported"; 86 $prefix=`echo __USER_LABEL_PREFIX__ | $ENV{CC} -E -P -`; 87 chomp($prefix); 88 } 89elsif ($flavour eq "macosx") { $gas=1; $elf=0; $prefix="_"; $decor="L\$"; } 90elsif ($flavour eq "masm") { $gas=0; $elf=0; $masm=$masmref; $win64=1; $decor="\$L\$"; } 91elsif ($flavour eq "nasm") { $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $decor="\$L\$"; $PTR=""; } 92elsif (!$gas) { die "unknown flavour $flavour"; } 93 94my $current_segment; 95my $current_function; 96my %globals; 97 98{ package opcode; # pick up opcodes 99 sub re { 100 my $self = shift; # single instance in enough... 101 local *line = shift; 102 undef $ret; 103 104 if ($line =~ /^([a-z][a-z0-9]*)/i) { 105 $self->{op} = $1; 106 $ret = $self; 107 $line = substr($line,@+[0]); $line =~ s/^\s+//; 108 109 undef $self->{sz}; 110 if ($self->{op} =~ /^(movz)x?([bw]).*/) { # movz is pain... 111 $self->{op} = $1; 112 $self->{sz} = $2; 113 } elsif ($self->{op} =~ /call|jmp/) { 114 $self->{sz} = ""; 115 } elsif ($self->{op} =~ /^p/ && $' !~ /^(ush|op|insrw)/) { # SSEn 116 $self->{sz} = ""; 117 } elsif ($self->{op} =~ /^v/) { # VEX 118 $self->{sz} = ""; 119 } elsif ($self->{op} =~ /mov[dq]/ && $line =~ /%xmm/) { 120 $self->{sz} = ""; 121 } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) { 122 $self->{op} = $1; 123 $self->{sz} = $2; 124 } 125 } 126 $ret; 127 } 128 sub size { 129 my $self = shift; 130 my $sz = shift; 131 $self->{sz} = $sz if (defined($sz) && !defined($self->{sz})); 132 $self->{sz}; 133 } 134 sub out { 135 my $self = shift; 136 if ($gas) { 137 if ($self->{op} eq "movz") { # movz is pain... 138 sprintf "%s%s%s",$self->{op},$self->{sz},shift; 139 } elsif ($self->{op} =~ /^set/) { 140 "$self->{op}"; 141 } elsif ($self->{op} eq "ret") { 142 my $epilogue = ""; 143 if ($win64 && $current_function->{abi} eq "svr4") { 144 $epilogue = "movq 8(%rsp),%rdi\n\t" . 145 "movq 16(%rsp),%rsi\n\t"; 146 } 147 $epilogue . ".byte 0xf3,0xc3"; 148 } elsif ($self->{op} eq "call" && !$elf && $current_segment eq ".init") { 149 ".p2align\t3\n\t.quad"; 150 } else { 151 "$self->{op}$self->{sz}"; 152 } 153 } else { 154 $self->{op} =~ s/^movz/movzx/; 155 if ($self->{op} eq "ret") { 156 $self->{op} = ""; 157 if ($win64 && $current_function->{abi} eq "svr4") { 158 $self->{op} = "mov rdi,QWORD${PTR}[8+rsp]\t;WIN64 epilogue\n\t". 159 "mov rsi,QWORD${PTR}[16+rsp]\n\t"; 160 } 161 $self->{op} .= "DB\t0F3h,0C3h\t\t;repret"; 162 } elsif ($self->{op} =~ /^(pop|push)f/) { 163 $self->{op} .= $self->{sz}; 164 } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") { 165 $self->{op} = "\tDQ"; 166 } 167 $self->{op}; 168 } 169 } 170 sub mnemonic { 171 my $self=shift; 172 my $op=shift; 173 $self->{op}=$op if (defined($op)); 174 $self->{op}; 175 } 176} 177{ package const; # pick up constants, which start with $ 178 sub re { 179 my $self = shift; # single instance in enough... 180 local *line = shift; 181 undef $ret; 182 183 if ($line =~ /^\$([^,]+)/) { 184 $self->{value} = $1; 185 $ret = $self; 186 $line = substr($line,@+[0]); $line =~ s/^\s+//; 187 } 188 $ret; 189 } 190 sub out { 191 my $self = shift; 192 193 if ($gas) { 194 # Solaris /usr/ccs/bin/as can't handle multiplications 195 # in $self->{value} 196 $self->{value} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi; 197 $self->{value} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg; 198 sprintf "\$%s",$self->{value}; 199 } else { 200 $self->{value} =~ s/(0b[0-1]+)/oct($1)/eig; 201 $self->{value} =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm); 202 sprintf "%s",$self->{value}; 203 } 204 } 205} 206{ package ea; # pick up effective addresses: expr(%reg,%reg,scale) 207 sub re { 208 my $self = shift; # single instance in enough... 209 local *line = shift; 210 undef $ret; 211 212 # optional * ---vvv--- appears in indirect jmp/call 213 if ($line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)/) { 214 $self->{asterisk} = $1; 215 $self->{label} = $2; 216 ($self->{base},$self->{index},$self->{scale})=split(/,/,$3); 217 $self->{scale} = 1 if (!defined($self->{scale})); 218 $ret = $self; 219 $line = substr($line,@+[0]); $line =~ s/^\s+//; 220 221 if ($win64 && $self->{label} =~ s/\@GOTPCREL//) { 222 die if (opcode->mnemonic() ne "mov"); 223 opcode->mnemonic("lea"); 224 } 225 $self->{base} =~ s/^%//; 226 $self->{index} =~ s/^%// if (defined($self->{index})); 227 } 228 $ret; 229 } 230 sub size {} 231 sub out { 232 my $self = shift; 233 my $sz = shift; 234 235 $self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 236 $self->{label} =~ s/\.L/$decor/g; 237 238 # Silently convert all EAs to 64-bit. This is required for 239 # elder GNU assembler and results in more compact code, 240 # *but* most importantly AES module depends on this feature! 241 $self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; 242 $self->{base} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; 243 244 # Solaris /usr/ccs/bin/as can't handle multiplications 245 # in $self->{label}, new gas requires sign extension... 246 use integer; 247 $self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi; 248 $self->{label} =~ s/\b([0-9]+\s*[\*\/\%]\s*[0-9]+)\b/eval($1)/eg; 249 $self->{label} =~ s/\b([0-9]+)\b/$1<<32>>32/eg; 250 251 if (!$self->{label} && $self->{index} && $self->{scale}==1 && 252 $self->{base} =~ /(rbp|r13)/) { 253 $self->{base} = $self->{index}; $self->{index} = $1; 254 } 255 256 if ($gas) { 257 $self->{label} =~ s/^___imp_/__imp__/ if ($flavour eq "mingw64"); 258 259 if (defined($self->{index})) { 260 sprintf "%s%s(%s,%%%s,%d)",$self->{asterisk}, 261 $self->{label}, 262 $self->{base}?"%$self->{base}":"", 263 $self->{index},$self->{scale}; 264 } else { 265 sprintf "%s%s(%%%s)", $self->{asterisk},$self->{label},$self->{base}; 266 } 267 } else { 268 %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR", 269 l=>"DWORD$PTR", d=>"DWORD$PTR", 270 q=>"QWORD$PTR", o=>"OWORD$PTR", 271 x=>"XMMWORD$PTR", y=>"YMMWORD$PTR", z=>"ZMMWORD$PTR" ); 272 273 $self->{label} =~ s/\./\$/g; 274 $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig; 275 $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/); 276 277 ($self->{asterisk}) && ($sz="q") || 278 (opcode->mnemonic() =~ /^v?mov([qd])$/) && ($sz=$1) || 279 (opcode->mnemonic() =~ /^v?pinsr([qdwb])$/) && ($sz=$1) || 280 (opcode->mnemonic() =~ /^vpbroadcast([qdwb])$/) && ($sz=$1) || 281 (opcode->mnemonic() =~ /^vinsert[fi]128$/) && ($sz="x"); 282 283 if (defined($self->{index})) { 284 sprintf "%s[%s%s*%d%s]",$szmap{$sz}, 285 $self->{label}?"$self->{label}+":"", 286 $self->{index},$self->{scale}, 287 $self->{base}?"+$self->{base}":""; 288 } elsif ($self->{base} eq "rip") { 289 sprintf "%s[%s]",$szmap{$sz},$self->{label}; 290 } else { 291 sprintf "%s[%s%s]",$szmap{$sz}, 292 $self->{label}?"$self->{label}+":"", 293 $self->{base}; 294 } 295 } 296 } 297} 298{ package register; # pick up registers, which start with %. 299 sub re { 300 my $class = shift; # muliple instances... 301 my $self = {}; 302 local *line = shift; 303 undef $ret; 304 305 # optional * ---vvv--- appears in indirect jmp/call 306 if ($line =~ /^(\*?)%(\w+)/) { 307 bless $self,$class; 308 $self->{asterisk} = $1; 309 $self->{value} = $2; 310 $ret = $self; 311 $line = substr($line,@+[0]); $line =~ s/^\s+//; 312 } 313 $ret; 314 } 315 sub size { 316 my $self = shift; 317 undef $ret; 318 319 if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; } 320 elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; } 321 elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; } 322 elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; } 323 elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; } 324 elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; } 325 elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; } 326 elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; } 327 328 $ret; 329 } 330 sub out { 331 my $self = shift; 332 if ($gas) { sprintf "%s%%%s",$self->{asterisk},$self->{value}; } 333 else { $self->{value}; } 334 } 335} 336{ package label; # pick up labels, which end with : 337 sub re { 338 my $self = shift; # single instance is enough... 339 local *line = shift; 340 undef $ret; 341 342 if ($line =~ /(^[\.\w]+)\:/) { 343 $self->{value} = $1; 344 $ret = $self; 345 $line = substr($line,@+[0]); $line =~ s/^\s+//; 346 347 $self->{value} =~ s/^\.L/$decor/; 348 } 349 $ret; 350 } 351 sub out { 352 my $self = shift; 353 354 if ($gas) { 355 my $func = ($globals{$self->{value}} or $self->{value}) . ":"; 356 if ($win64 && 357 $current_function->{name} eq $self->{value} && 358 $current_function->{abi} eq "svr4") { 359 $func .= "\n"; 360 $func .= " movq %rdi,8(%rsp)\n"; 361 $func .= " movq %rsi,16(%rsp)\n"; 362 $func .= " movq %rsp,%rax\n"; 363 $func .= "${decor}SEH_begin_$current_function->{name}:\n"; 364 my $narg = $current_function->{narg}; 365 $narg=6 if (!defined($narg)); 366 $func .= " movq %rcx,%rdi\n" if ($narg>0); 367 $func .= " movq %rdx,%rsi\n" if ($narg>1); 368 $func .= " movq %r8,%rdx\n" if ($narg>2); 369 $func .= " movq %r9,%rcx\n" if ($narg>3); 370 $func .= " movq 40(%rsp),%r8\n" if ($narg>4); 371 $func .= " movq 48(%rsp),%r9\n" if ($narg>5); 372 } 373 $func; 374 } elsif ($self->{value} ne "$current_function->{name}") { 375 $self->{value} .= ":" if ($masm && $ret!~m/^\$/); 376 $self->{value} . ":"; 377 } elsif ($win64 && $current_function->{abi} eq "svr4") { 378 my $func = "$current_function->{name}" . 379 ($nasm ? ":" : "\tPROC $current_function->{scope}") . 380 "\n"; 381 $func .= " mov QWORD${PTR}[8+rsp],rdi\t;WIN64 prologue\n"; 382 $func .= " mov QWORD${PTR}[16+rsp],rsi\n"; 383 $func .= " mov rax,rsp\n"; 384 $func .= "${decor}SEH_begin_$current_function->{name}:"; 385 $func .= ":" if ($masm); 386 $func .= "\n"; 387 my $narg = $current_function->{narg}; 388 $narg=6 if (!defined($narg)); 389 $func .= " mov rdi,rcx\n" if ($narg>0); 390 $func .= " mov rsi,rdx\n" if ($narg>1); 391 $func .= " mov rdx,r8\n" if ($narg>2); 392 $func .= " mov rcx,r9\n" if ($narg>3); 393 $func .= " mov r8,QWORD${PTR}[40+rsp]\n" if ($narg>4); 394 $func .= " mov r9,QWORD${PTR}[48+rsp]\n" if ($narg>5); 395 $func .= "\n"; 396 } else { 397 "$current_function->{name}". 398 ($nasm ? ":" : "\tPROC $current_function->{scope}"); 399 } 400 } 401} 402{ package expr; # pick up expressioins 403 sub re { 404 my $self = shift; # single instance is enough... 405 local *line = shift; 406 undef $ret; 407 408 if ($line =~ /(^[^,]+)/) { 409 $self->{value} = $1; 410 $ret = $self; 411 $line = substr($line,@+[0]); $line =~ s/^\s+//; 412 413 $self->{value} =~ s/\@PLT// if (!$elf); 414 $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 415 $self->{value} =~ s/\.L/$decor/g; 416 } 417 $ret; 418 } 419 sub out { 420 my $self = shift; 421 if ($nasm && opcode->mnemonic()=~m/^j(?![re]cxz)/) { 422 "NEAR ".$self->{value}; 423 } else { 424 $self->{value}; 425 } 426 } 427} 428{ package directive; # pick up directives, which start with . 429 sub re { 430 my $self = shift; # single instance is enough... 431 local *line = shift; 432 undef $ret; 433 my $dir; 434 my %opcode = # lea 2f-1f(%rip),%dst; 1: nop; 2: 435 ( "%rax"=>0x01058d48, "%rcx"=>0x010d8d48, 436 "%rdx"=>0x01158d48, "%rbx"=>0x011d8d48, 437 "%rsp"=>0x01258d48, "%rbp"=>0x012d8d48, 438 "%rsi"=>0x01358d48, "%rdi"=>0x013d8d48, 439 "%r8" =>0x01058d4c, "%r9" =>0x010d8d4c, 440 "%r10"=>0x01158d4c, "%r11"=>0x011d8d4c, 441 "%r12"=>0x01258d4c, "%r13"=>0x012d8d4c, 442 "%r14"=>0x01358d4c, "%r15"=>0x013d8d4c ); 443 444 if ($line =~ /^\s*(\.\w+)/) { 445 $dir = $1; 446 $ret = $self; 447 undef $self->{value}; 448 $line = substr($line,@+[0]); $line =~ s/^\s+//; 449 450 SWITCH: for ($dir) { 451 /\.picmeup/ && do { if ($line =~ /(%r[\w]+)/i) { 452 $dir="\t.long"; 453 $line=sprintf "0x%x,0x90000000",$opcode{$1}; 454 } 455 last; 456 }; 457 /\.global|\.globl|\.extern/ 458 && do { $globals{$line} = $prefix . $line; 459 $line = $globals{$line} if ($prefix); 460 last; 461 }; 462 /\.type/ && do { ($sym,$type,$narg) = split(',',$line); 463 if ($type eq "\@function") { 464 undef $current_function; 465 $current_function->{name} = $sym; 466 $current_function->{abi} = "svr4"; 467 $current_function->{narg} = $narg; 468 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 469 } elsif ($type eq "\@abi-omnipotent") { 470 undef $current_function; 471 $current_function->{name} = $sym; 472 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 473 } 474 $line =~ s/\@abi\-omnipotent/\@function/; 475 $line =~ s/\@function.*/\@function/; 476 last; 477 }; 478 /\.asciz/ && do { if ($line =~ /^"(.*)"$/) { 479 $dir = ".byte"; 480 $line = join(",",unpack("C*",$1),0); 481 } 482 last; 483 }; 484 /\.rva|\.long|\.quad/ 485 && do { $line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 486 $line =~ s/\.L/$decor/g; 487 last; 488 }; 489 } 490 491 if ($gas) { 492 $self->{value} = $dir . "\t" . $line; 493 494 if ($dir =~ /\.extern/) { 495 if ($flavour eq "elf") { 496 $self->{value} .= "\n.hidden $line"; 497 } else { 498 $self->{value} = ""; 499 } 500 } elsif (!$elf && $dir =~ /\.type/) { 501 $self->{value} = ""; 502 $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" . 503 (defined($globals{$1})?".scl 2;":".scl 3;") . 504 "\t.type 32;\t.endef" 505 if ($win64 && $line =~ /([^,]+),\@function/); 506 } elsif (!$elf && $dir =~ /\.size/) { 507 $self->{value} = ""; 508 if (defined($current_function)) { 509 $self->{value} .= "${decor}SEH_end_$current_function->{name}:" 510 if ($win64 && $current_function->{abi} eq "svr4"); 511 undef $current_function; 512 } 513 } elsif (!$elf && $dir =~ /\.align/) { 514 $self->{value} = ".p2align\t" . (log($line)/log(2)); 515 } elsif ($dir eq ".section") { 516 $current_segment=$line; 517 if (!$elf && $current_segment eq ".init") { 518 if ($flavour eq "macosx") { $self->{value} = ".mod_init_func"; } 519 elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.ctors"; } 520 } 521 } elsif ($dir =~ /\.(text|data)/) { 522 $current_segment=".$1"; 523 } elsif ($dir =~ /\.global|\.globl|\.extern/) { 524 if ($flavour eq "macosx") { 525 $self->{value} .= "\n.private_extern $line"; 526 } else { 527 $self->{value} .= "\n.hidden $line"; 528 } 529 } elsif ($dir =~ /\.hidden/) { 530 if ($flavour eq "macosx") { $self->{value} = ".private_extern\t$prefix$line"; } 531 elsif ($flavour eq "mingw64") { $self->{value} = ""; } 532 } elsif ($dir =~ /\.comm/) { 533 $self->{value} = "$dir\t$prefix$line"; 534 $self->{value} =~ s|,([0-9]+),([0-9]+)$|",$1,".log($2)/log(2)|e if ($flavour eq "macosx"); 535 } 536 $line = ""; 537 return $self; 538 } 539 540 # non-gas case or nasm/masm 541 SWITCH: for ($dir) { 542 /\.text/ && do { my $v=undef; 543 if ($nasm) { 544 $v="section .text code align=64\n"; 545 } else { 546 $v="$current_segment\tENDS\n" if ($current_segment); 547 $current_segment = ".text\$"; 548 $v.="$current_segment\tSEGMENT "; 549 $v.=$masm>=$masmref ? "ALIGN(256)" : "PAGE"; 550 $v.=" 'CODE'"; 551 } 552 $self->{value} = $v; 553 last; 554 }; 555 /\.data/ && do { my $v=undef; 556 if ($nasm) { 557 $v="section .data data align=8\n"; 558 } else { 559 $v="$current_segment\tENDS\n" if ($current_segment); 560 $current_segment = "_DATA"; 561 $v.="$current_segment\tSEGMENT"; 562 } 563 $self->{value} = $v; 564 last; 565 }; 566 /\.section/ && do { my $v=undef; 567 $line =~ s/([^,]*).*/$1/; 568 $line = ".CRT\$XCU" if ($line eq ".init"); 569 if ($nasm) { 570 $v="section $line"; 571 if ($line=~/\.([px])data/) { 572 $v.=" rdata align="; 573 $v.=$1 eq "p"? 4 : 8; 574 } elsif ($line=~/\.CRT\$/i) { 575 $v.=" rdata align=8"; 576 } 577 } else { 578 $v="$current_segment\tENDS\n" if ($current_segment); 579 $v.="$line\tSEGMENT"; 580 if ($line=~/\.([px])data/) { 581 $v.=" READONLY"; 582 $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref); 583 } elsif ($line=~/\.CRT\$/i) { 584 $v.=" READONLY "; 585 $v.=$masm>=$masmref ? "ALIGN(8)" : "DWORD"; 586 } 587 } 588 $current_segment = $line; 589 $self->{value} = $v; 590 last; 591 }; 592 /\.extern/ && do { $self->{value} = "EXTERN\t".$line; 593 $self->{value} .= ":NEAR" if ($masm); 594 last; 595 }; 596 /\.globl|.global/ 597 && do { $self->{value} = $masm?"PUBLIC":"global"; 598 $self->{value} .= "\t".$line; 599 last; 600 }; 601 /\.size/ && do { if (defined($current_function)) { 602 undef $self->{value}; 603 if ($current_function->{abi} eq "svr4") { 604 $self->{value}="${decor}SEH_end_$current_function->{name}:"; 605 $self->{value}.=":\n" if($masm); 606 } 607 $self->{value}.="$current_function->{name}\tENDP" if($masm && $current_function->{name}); 608 undef $current_function; 609 } 610 last; 611 }; 612 /\.align/ && do { $self->{value} = "ALIGN\t".$line; last; }; 613 /\.(value|long|rva|quad)/ 614 && do { my $sz = substr($1,0,1); 615 my @arr = split(/,\s*/,$line); 616 my $last = pop(@arr); 617 my $conv = sub { my $var=shift; 618 $var=~s/^(0b[0-1]+)/oct($1)/eig; 619 $var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm); 620 if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva")) 621 { $var=~s/([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; } 622 $var; 623 }; 624 625 $sz =~ tr/bvlrq/BWDDQ/; 626 $self->{value} = "\tD$sz\t"; 627 for (@arr) { $self->{value} .= &$conv($_).","; } 628 $self->{value} .= &$conv($last); 629 last; 630 }; 631 /\.byte/ && do { my @str=split(/,\s*/,$line); 632 map(s/(0b[0-1]+)/oct($1)/eig,@str); 633 map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm); 634 while ($#str>15) { 635 $self->{value}.="DB\t" 636 .join(",",@str[0..15])."\n"; 637 foreach (0..15) { shift @str; } 638 } 639 $self->{value}.="DB\t" 640 .join(",",@str) if (@str); 641 last; 642 }; 643 /\.comm/ && do { my @str=split(/,\s*/,$line); 644 my $v=undef; 645 if ($nasm) { 646 $v.="common $prefix@str[0] @str[1]"; 647 } else { 648 $v="$current_segment\tENDS\n" if ($current_segment); 649 $current_segment = "_DATA"; 650 $v.="$current_segment\tSEGMENT\n"; 651 $v.="COMM @str[0]:DWORD:".@str[1]/4; 652 } 653 $self->{value} = $v; 654 last; 655 }; 656 } 657 $line = ""; 658 } 659 660 $ret; 661 } 662 sub out { 663 my $self = shift; 664 $self->{value}; 665 } 666} 667 668sub rex { 669 local *opcode=shift; 670 my ($dst,$src,$rex)=@_; 671 672 $rex|=0x04 if($dst>=8); 673 $rex|=0x01 if($src>=8); 674 push @opcode,($rex|0x40) if ($rex); 675} 676 677# older gas and ml64 don't handle SSE>2 instructions 678my %regrm = ( "%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3, 679 "%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7 ); 680 681my $movq = sub { # elderly gas can't handle inter-register movq 682 my $arg = shift; 683 my @opcode=(0x66); 684 if ($arg =~ /%xmm([0-9]+),\s*%r(\w+)/) { 685 my ($src,$dst)=($1,$2); 686 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 687 rex(\@opcode,$src,$dst,0x8); 688 push @opcode,0x0f,0x7e; 689 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 690 @opcode; 691 } elsif ($arg =~ /%r(\w+),\s*%xmm([0-9]+)/) { 692 my ($src,$dst)=($2,$1); 693 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 694 rex(\@opcode,$src,$dst,0x8); 695 push @opcode,0x0f,0x6e; 696 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 697 @opcode; 698 } else { 699 (); 700 } 701}; 702 703my $pextrd = sub { 704 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*(%\w+)/) { 705 my @opcode=(0x66); 706 $imm=$1; 707 $src=$2; 708 $dst=$3; 709 if ($dst =~ /%r([0-9]+)d/) { $dst = $1; } 710 elsif ($dst =~ /%e/) { $dst = $regrm{$dst}; } 711 rex(\@opcode,$src,$dst); 712 push @opcode,0x0f,0x3a,0x16; 713 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 714 push @opcode,$imm; 715 @opcode; 716 } else { 717 (); 718 } 719}; 720 721my $pinsrd = sub { 722 if (shift =~ /\$([0-9]+),\s*(%\w+),\s*%xmm([0-9]+)/) { 723 my @opcode=(0x66); 724 $imm=$1; 725 $src=$2; 726 $dst=$3; 727 if ($src =~ /%r([0-9]+)/) { $src = $1; } 728 elsif ($src =~ /%e/) { $src = $regrm{$src}; } 729 rex(\@opcode,$dst,$src); 730 push @opcode,0x0f,0x3a,0x22; 731 push @opcode,0xc0|(($dst&7)<<3)|($src&7); # ModR/M 732 push @opcode,$imm; 733 @opcode; 734 } else { 735 (); 736 } 737}; 738 739my $pshufb = sub { 740 if (shift =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) { 741 my @opcode=(0x66); 742 rex(\@opcode,$2,$1); 743 push @opcode,0x0f,0x38,0x00; 744 push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M 745 @opcode; 746 } else { 747 (); 748 } 749}; 750 751my $palignr = sub { 752 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 753 my @opcode=(0x66); 754 rex(\@opcode,$3,$2); 755 push @opcode,0x0f,0x3a,0x0f; 756 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 757 push @opcode,$1; 758 @opcode; 759 } else { 760 (); 761 } 762}; 763 764my $pclmulqdq = sub { 765 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 766 my @opcode=(0x66); 767 rex(\@opcode,$3,$2); 768 push @opcode,0x0f,0x3a,0x44; 769 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 770 my $c=$1; 771 push @opcode,$c=~/^0/?oct($c):$c; 772 @opcode; 773 } else { 774 (); 775 } 776}; 777 778my $rdrand = sub { 779 if (shift =~ /%[er](\w+)/) { 780 my @opcode=(); 781 my $dst=$1; 782 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 783 rex(\@opcode,0,$1,8); 784 push @opcode,0x0f,0xc7,0xf0|($dst&7); 785 @opcode; 786 } else { 787 (); 788 } 789}; 790 791my $rdseed = sub { 792 if (shift =~ /%[er](\w+)/) { 793 my @opcode=(); 794 my $dst=$1; 795 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 796 rex(\@opcode,0,$1,8); 797 push @opcode,0x0f,0xc7,0xf8|($dst&7); 798 @opcode; 799 } else { 800 (); 801 } 802}; 803 804sub rxb { 805 local *opcode=shift; 806 my ($dst,$src1,$src2,$rxb)=@_; 807 808 $rxb|=0x7<<5; 809 $rxb&=~(0x04<<5) if($dst>=8); 810 $rxb&=~(0x01<<5) if($src1>=8); 811 $rxb&=~(0x02<<5) if($src2>=8); 812 push @opcode,$rxb; 813} 814 815my $vprotd = sub { 816 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 817 my @opcode=(0x8f); 818 rxb(\@opcode,$3,$2,-1,0x08); 819 push @opcode,0x78,0xc2; 820 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 821 my $c=$1; 822 push @opcode,$c=~/^0/?oct($c):$c; 823 @opcode; 824 } else { 825 (); 826 } 827}; 828 829my $vprotq = sub { 830 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 831 my @opcode=(0x8f); 832 rxb(\@opcode,$3,$2,-1,0x08); 833 push @opcode,0x78,0xc3; 834 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 835 my $c=$1; 836 push @opcode,$c=~/^0/?oct($c):$c; 837 @opcode; 838 } else { 839 (); 840 } 841}; 842 843if ($nasm) { 844 print <<___; 845default rel 846%define XMMWORD 847%define YMMWORD 848%define ZMMWORD 849___ 850} elsif ($masm) { 851 print <<___; 852OPTION DOTNAME 853___ 854} 855 856print STDOUT "#if defined(__x86_64__)\n" if ($gas); 857 858while($line=<>) { 859 860 chomp($line); 861 862 $line =~ s|[#!].*$||; # get rid of asm-style comments... 863 $line =~ s|/\*.*\*/||; # ... and C-style comments... 864 $line =~ s|^\s+||; # ... and skip white spaces in beginning 865 $line =~ s|\s+$||; # ... and at the end 866 867 undef $label; 868 undef $opcode; 869 undef @args; 870 871 if ($label=label->re(\$line)) { print $label->out(); } 872 873 if (directive->re(\$line)) { 874 printf "%s",directive->out(); 875 } elsif ($opcode=opcode->re(\$line)) { 876 my $asm = eval("\$".$opcode->mnemonic()); 877 undef @bytes; 878 879 if ((ref($asm) eq 'CODE') && scalar(@bytes=&$asm($line))) { 880 print $gas?".byte\t":"DB\t",join(',',@bytes),"\n"; 881 next; 882 } 883 884 ARGUMENT: while (1) { 885 my $arg; 886 887 if ($arg=register->re(\$line)) { opcode->size($arg->size()); } 888 elsif ($arg=const->re(\$line)) { } 889 elsif ($arg=ea->re(\$line)) { } 890 elsif ($arg=expr->re(\$line)) { } 891 else { last ARGUMENT; } 892 893 push @args,$arg; 894 895 last ARGUMENT if ($line !~ /^,/); 896 897 $line =~ s/^,\s*//; 898 } # ARGUMENT: 899 900 if ($#args>=0) { 901 my $insn; 902 my $sz=opcode->size(); 903 904 if ($gas) { 905 $insn = $opcode->out($#args>=1?$args[$#args]->size():$sz); 906 @args = map($_->out($sz),@args); 907 printf "\t%s\t%s",$insn,join(",",@args); 908 } else { 909 $insn = $opcode->out(); 910 foreach (@args) { 911 my $arg = $_->out(); 912 # $insn.=$sz compensates for movq, pinsrw, ... 913 if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; } 914 if ($arg =~ /^ymm[0-9]+$/) { $insn.=$sz; $sz="y" if(!$sz); last; } 915 if ($arg =~ /^zmm[0-9]+$/) { $insn.=$sz; $sz="z" if(!$sz); last; } 916 if ($arg =~ /^mm[0-9]+$/) { $insn.=$sz; $sz="q" if(!$sz); last; } 917 } 918 @args = reverse(@args); 919 undef $sz if ($nasm && $opcode->mnemonic() eq "lea"); 920 921 if ($insn eq "movq" && $#args == 1 && $args[0]->out($sz) eq "xmm0" && $args[1]->out($sz) eq "rax") { 922 # I have no clue why MASM can't parse this instruction. 923 printf "DB 66h, 48h, 0fh, 6eh, 0c0h"; 924 } else { 925 printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args)); 926 } 927 } 928 } else { 929 printf "\t%s",$opcode->out(); 930 } 931 } 932 933 print $line,"\n"; 934} 935 936print "\n$current_segment\tENDS\n" if ($current_segment && $masm); 937print "END\n" if ($masm); 938print "#endif\n" if ($gas); 939 940 941close STDOUT; 942 943 ################################################# 944# Cross-reference x86_64 ABI "card" 945# 946# Unix Win64 947# %rax * * 948# %rbx - - 949# %rcx #4 #1 950# %rdx #3 #2 951# %rsi #2 - 952# %rdi #1 - 953# %rbp - - 954# %rsp - - 955# %r8 #5 #3 956# %r9 #6 #4 957# %r10 * * 958# %r11 * * 959# %r12 - - 960# %r13 - - 961# %r14 - - 962# %r15 - - 963# 964# (*) volatile register 965# (-) preserved by callee 966# (#) Nth argument, volatile 967# 968# In Unix terms top of stack is argument transfer area for arguments 969# which could not be accomodated in registers. Or in other words 7th 970# [integer] argument resides at 8(%rsp) upon function entry point. 971# 128 bytes above %rsp constitute a "red zone" which is not touched 972# by signal handlers and can be used as temporal storage without 973# allocating a frame. 974# 975# In Win64 terms N*8 bytes on top of stack is argument transfer area, 976# which belongs to/can be overwritten by callee. N is the number of 977# arguments passed to callee, *but* not less than 4! This means that 978# upon function entry point 5th argument resides at 40(%rsp), as well 979# as that 32 bytes from 8(%rsp) can always be used as temporal 980# storage [without allocating a frame]. One can actually argue that 981# one can assume a "red zone" above stack pointer under Win64 as well. 982# Point is that at apparently no occasion Windows kernel would alter 983# the area above user stack pointer in true asynchronous manner... 984# 985# All the above means that if assembler programmer adheres to Unix 986# register and stack layout, but disregards the "red zone" existense, 987# it's possible to use following prologue and epilogue to "gear" from 988# Unix to Win64 ABI in leaf functions with not more than 6 arguments. 989# 990# omnipotent_function: 991# ifdef WIN64 992# movq %rdi,8(%rsp) 993# movq %rsi,16(%rsp) 994# movq %rcx,%rdi ; if 1st argument is actually present 995# movq %rdx,%rsi ; if 2nd argument is actually ... 996# movq %r8,%rdx ; if 3rd argument is ... 997# movq %r9,%rcx ; if 4th argument ... 998# movq 40(%rsp),%r8 ; if 5th ... 999# movq 48(%rsp),%r9 ; if 6th ... 1000# endif 1001# ... 1002# ifdef WIN64 1003# movq 8(%rsp),%rdi 1004# movq 16(%rsp),%rsi 1005# endif 1006# ret 1007# 1008 ################################################# 1009# Win64 SEH, Structured Exception Handling. 1010# 1011# Unlike on Unix systems(*) lack of Win64 stack unwinding information 1012# has undesired side-effect at run-time: if an exception is raised in 1013# assembler subroutine such as those in question (basically we're 1014# referring to segmentation violations caused by malformed input 1015# parameters), the application is briskly terminated without invoking 1016# any exception handlers, most notably without generating memory dump 1017# or any user notification whatsoever. This poses a problem. It's 1018# possible to address it by registering custom language-specific 1019# handler that would restore processor context to the state at 1020# subroutine entry point and return "exception is not handled, keep 1021# unwinding" code. Writing such handler can be a challenge... But it's 1022# doable, though requires certain coding convention. Consider following 1023# snippet: 1024# 1025# .type function,@function 1026# function: 1027# movq %rsp,%rax # copy rsp to volatile register 1028# pushq %r15 # save non-volatile registers 1029# pushq %rbx 1030# pushq %rbp 1031# movq %rsp,%r11 1032# subq %rdi,%r11 # prepare [variable] stack frame 1033# andq $-64,%r11 1034# movq %rax,0(%r11) # check for exceptions 1035# movq %r11,%rsp # allocate [variable] stack frame 1036# movq %rax,0(%rsp) # save original rsp value 1037# magic_point: 1038# ... 1039# movq 0(%rsp),%rcx # pull original rsp value 1040# movq -24(%rcx),%rbp # restore non-volatile registers 1041# movq -16(%rcx),%rbx 1042# movq -8(%rcx),%r15 1043# movq %rcx,%rsp # restore original rsp 1044# ret 1045# .size function,.-function 1046# 1047# The key is that up to magic_point copy of original rsp value remains 1048# in chosen volatile register and no non-volatile register, except for 1049# rsp, is modified. While past magic_point rsp remains constant till 1050# the very end of the function. In this case custom language-specific 1051# exception handler would look like this: 1052# 1053# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, 1054# CONTEXT *context,DISPATCHER_CONTEXT *disp) 1055# { ULONG64 *rsp = (ULONG64 *)context->Rax; 1056# if (context->Rip >= magic_point) 1057# { rsp = ((ULONG64 **)context->Rsp)[0]; 1058# context->Rbp = rsp[-3]; 1059# context->Rbx = rsp[-2]; 1060# context->R15 = rsp[-1]; 1061# } 1062# context->Rsp = (ULONG64)rsp; 1063# context->Rdi = rsp[1]; 1064# context->Rsi = rsp[2]; 1065# 1066# memcpy (disp->ContextRecord,context,sizeof(CONTEXT)); 1067# RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase, 1068# dips->ControlPc,disp->FunctionEntry,disp->ContextRecord, 1069# &disp->HandlerData,&disp->EstablisherFrame,NULL); 1070# return ExceptionContinueSearch; 1071# } 1072# 1073# It's appropriate to implement this handler in assembler, directly in 1074# function's module. In order to do that one has to know members' 1075# offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant 1076# values. Here they are: 1077# 1078# CONTEXT.Rax 120 1079# CONTEXT.Rcx 128 1080# CONTEXT.Rdx 136 1081# CONTEXT.Rbx 144 1082# CONTEXT.Rsp 152 1083# CONTEXT.Rbp 160 1084# CONTEXT.Rsi 168 1085# CONTEXT.Rdi 176 1086# CONTEXT.R8 184 1087# CONTEXT.R9 192 1088# CONTEXT.R10 200 1089# CONTEXT.R11 208 1090# CONTEXT.R12 216 1091# CONTEXT.R13 224 1092# CONTEXT.R14 232 1093# CONTEXT.R15 240 1094# CONTEXT.Rip 248 1095# CONTEXT.Xmm6 512 1096# sizeof(CONTEXT) 1232 1097# DISPATCHER_CONTEXT.ControlPc 0 1098# DISPATCHER_CONTEXT.ImageBase 8 1099# DISPATCHER_CONTEXT.FunctionEntry 16 1100# DISPATCHER_CONTEXT.EstablisherFrame 24 1101# DISPATCHER_CONTEXT.TargetIp 32 1102# DISPATCHER_CONTEXT.ContextRecord 40 1103# DISPATCHER_CONTEXT.LanguageHandler 48 1104# DISPATCHER_CONTEXT.HandlerData 56 1105# UNW_FLAG_NHANDLER 0 1106# ExceptionContinueSearch 1 1107# 1108# In order to tie the handler to the function one has to compose 1109# couple of structures: one for .xdata segment and one for .pdata. 1110# 1111# UNWIND_INFO structure for .xdata segment would be 1112# 1113# function_unwind_info: 1114# .byte 9,0,0,0 1115# .rva handler 1116# 1117# This structure designates exception handler for a function with 1118# zero-length prologue, no stack frame or frame register. 1119# 1120# To facilitate composing of .pdata structures, auto-generated "gear" 1121# prologue copies rsp value to rax and denotes next instruction with 1122# .LSEH_begin_{function_name} label. This essentially defines the SEH 1123# styling rule mentioned in the beginning. Position of this label is 1124# chosen in such manner that possible exceptions raised in the "gear" 1125# prologue would be accounted to caller and unwound from latter's frame. 1126# End of function is marked with respective .LSEH_end_{function_name} 1127# label. To summarize, .pdata segment would contain 1128# 1129# .rva .LSEH_begin_function 1130# .rva .LSEH_end_function 1131# .rva function_unwind_info 1132# 1133# Reference to functon_unwind_info from .xdata segment is the anchor. 1134# In case you wonder why references are 32-bit .rvas and not 64-bit 1135# .quads. References put into these two segments are required to be 1136# *relative* to the base address of the current binary module, a.k.a. 1137# image base. No Win64 module, be it .exe or .dll, can be larger than 1138# 2GB and thus such relative references can be and are accommodated in 1139# 32 bits. 1140# 1141# Having reviewed the example function code, one can argue that "movq 1142# %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix 1143# rax would contain an undefined value. If this "offends" you, use 1144# another register and refrain from modifying rax till magic_point is 1145# reached, i.e. as if it was a non-volatile register. If more registers 1146# are required prior [variable] frame setup is completed, note that 1147# nobody says that you can have only one "magic point." You can 1148# "liberate" non-volatile registers by denoting last stack off-load 1149# instruction and reflecting it in finer grade unwind logic in handler. 1150# After all, isn't it why it's called *language-specific* handler... 1151# 1152# Attentive reader can notice that exceptions would be mishandled in 1153# auto-generated "gear" epilogue. Well, exception effectively can't 1154# occur there, because if memory area used by it was subject to 1155# segmentation violation, then it would be raised upon call to the 1156# function (and as already mentioned be accounted to caller, which is 1157# not a problem). If you're still not comfortable, then define tail 1158# "magic point" just prior ret instruction and have handler treat it... 1159# 1160# (*) Note that we're talking about run-time, not debug-time. Lack of 1161# unwind information makes debugging hard on both Windows and 1162# Unix. "Unlike" referes to the fact that on Unix signal handler 1163# will always be invoked, core dumped and appropriate exit code 1164# returned to parent (for user notification). 1165