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