x86_64-xlate.pl revision 325335
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 my $value = $self->{value}; 210 $value =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm); 211 sprintf "%s",$value; 212 } 213 } 214} 215{ package ea; # pick up effective addresses: expr(%reg,%reg,scale) 216 sub re { 217 my $self = shift; # single instance in enough... 218 local *line = shift; 219 undef $ret; 220 221 # optional * ---vvv--- appears in indirect jmp/call 222 if ($line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)/) { 223 $self->{asterisk} = $1; 224 $self->{label} = $2; 225 ($self->{base},$self->{index},$self->{scale})=split(/,/,$3); 226 $self->{scale} = 1 if (!defined($self->{scale})); 227 $ret = $self; 228 $line = substr($line,@+[0]); $line =~ s/^\s+//; 229 230 if ($win64 && $self->{label} =~ s/\@GOTPCREL//) { 231 die if (opcode->mnemonic() ne "mov"); 232 opcode->mnemonic("lea"); 233 } 234 $self->{base} =~ s/^%//; 235 $self->{index} =~ s/^%// if (defined($self->{index})); 236 } 237 $ret; 238 } 239 sub size {} 240 sub out { 241 my $self = shift; 242 my $sz = shift; 243 244 $self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 245 $self->{label} =~ s/\.L/$decor/g; 246 247 # Silently convert all EAs to 64-bit. This is required for 248 # elder GNU assembler and results in more compact code, 249 # *but* most importantly AES module depends on this feature! 250 $self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; 251 $self->{base} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; 252 253 # Solaris /usr/ccs/bin/as can't handle multiplications 254 # in $self->{label}... 255 use integer; 256 $self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi; 257 $self->{label} =~ s/\b([0-9]+\s*[\*\/\%]\s*[0-9]+)\b/eval($1)/eg; 258 259 # Some assemblers insist on signed presentation of 32-bit 260 # offsets, but sign extension is a tricky business in perl... 261 if ((1<<31)<<1) { 262 $self->{label} =~ s/\b([0-9]+)\b/$1<<32>>32/eg; 263 } else { 264 $self->{label} =~ s/\b([0-9]+)\b/$1>>0/eg; 265 } 266 267 if (!$self->{label} && $self->{index} && $self->{scale}==1 && 268 $self->{base} =~ /(rbp|r13)/) { 269 $self->{base} = $self->{index}; $self->{index} = $1; 270 } 271 272 if ($gas) { 273 $self->{label} =~ s/^___imp_/__imp__/ if ($flavour eq "mingw64"); 274 275 if (defined($self->{index})) { 276 sprintf "%s%s(%s,%%%s,%d)",$self->{asterisk}, 277 $self->{label}, 278 $self->{base}?"%$self->{base}":"", 279 $self->{index},$self->{scale}; 280 } else { 281 sprintf "%s%s(%%%s)", $self->{asterisk},$self->{label},$self->{base}; 282 } 283 } else { 284 %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR", 285 l=>"DWORD$PTR", d=>"DWORD$PTR", 286 q=>"QWORD$PTR", o=>"OWORD$PTR", 287 x=>"XMMWORD$PTR", y=>"YMMWORD$PTR", z=>"ZMMWORD$PTR" ); 288 289 $self->{label} =~ s/\./\$/g; 290 $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig; 291 $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/); 292 293 ($self->{asterisk}) && ($sz="q") || 294 (opcode->mnemonic() =~ /^v?mov([qd])$/) && ($sz=$1) || 295 (opcode->mnemonic() =~ /^v?pinsr([qdwb])$/) && ($sz=$1) || 296 (opcode->mnemonic() =~ /^vpbroadcast([qdwb])$/) && ($sz=$1) || 297 (opcode->mnemonic() =~ /^vinsert[fi]128$/) && ($sz="x"); 298 299 if (defined($self->{index})) { 300 sprintf "%s[%s%s*%d%s]",$szmap{$sz}, 301 $self->{label}?"$self->{label}+":"", 302 $self->{index},$self->{scale}, 303 $self->{base}?"+$self->{base}":""; 304 } elsif ($self->{base} eq "rip") { 305 sprintf "%s[%s]",$szmap{$sz},$self->{label}; 306 } else { 307 sprintf "%s[%s%s]",$szmap{$sz}, 308 $self->{label}?"$self->{label}+":"", 309 $self->{base}; 310 } 311 } 312 } 313} 314{ package register; # pick up registers, which start with %. 315 sub re { 316 my $class = shift; # muliple instances... 317 my $self = {}; 318 local *line = shift; 319 undef $ret; 320 321 # optional * ---vvv--- appears in indirect jmp/call 322 if ($line =~ /^(\*?)%(\w+)/) { 323 bless $self,$class; 324 $self->{asterisk} = $1; 325 $self->{value} = $2; 326 $ret = $self; 327 $line = substr($line,@+[0]); $line =~ s/^\s+//; 328 } 329 $ret; 330 } 331 sub size { 332 my $self = shift; 333 undef $ret; 334 335 if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; } 336 elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; } 337 elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; } 338 elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; } 339 elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; } 340 elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; } 341 elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; } 342 elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; } 343 344 $ret; 345 } 346 sub out { 347 my $self = shift; 348 if ($gas) { sprintf "%s%%%s",$self->{asterisk},$self->{value}; } 349 else { $self->{value}; } 350 } 351} 352{ package label; # pick up labels, which end with : 353 sub re { 354 my $self = shift; # single instance is enough... 355 local *line = shift; 356 undef $ret; 357 358 if ($line =~ /(^[\.\w]+)\:/) { 359 $self->{value} = $1; 360 $ret = $self; 361 $line = substr($line,@+[0]); $line =~ s/^\s+//; 362 363 $self->{value} =~ s/^\.L/$decor/; 364 } 365 $ret; 366 } 367 sub out { 368 my $self = shift; 369 370 if ($gas) { 371 my $func = ($globals{$self->{value}} or $self->{value}) . ":"; 372 if ($win64 && 373 $current_function->{name} eq $self->{value} && 374 $current_function->{abi} eq "svr4") { 375 $func .= "\n"; 376 $func .= " movq %rdi,8(%rsp)\n"; 377 $func .= " movq %rsi,16(%rsp)\n"; 378 $func .= " movq %rsp,%rax\n"; 379 $func .= "${decor}SEH_begin_$current_function->{name}:\n"; 380 my $narg = $current_function->{narg}; 381 $narg=6 if (!defined($narg)); 382 $func .= " movq %rcx,%rdi\n" if ($narg>0); 383 $func .= " movq %rdx,%rsi\n" if ($narg>1); 384 $func .= " movq %r8,%rdx\n" if ($narg>2); 385 $func .= " movq %r9,%rcx\n" if ($narg>3); 386 $func .= " movq 40(%rsp),%r8\n" if ($narg>4); 387 $func .= " movq 48(%rsp),%r9\n" if ($narg>5); 388 } 389 $func; 390 } elsif ($self->{value} ne "$current_function->{name}") { 391 $self->{value} .= ":" if ($masm && $ret!~m/^\$/); 392 $self->{value} . ":"; 393 } elsif ($win64 && $current_function->{abi} eq "svr4") { 394 my $func = "$current_function->{name}" . 395 ($nasm ? ":" : "\tPROC $current_function->{scope}") . 396 "\n"; 397 $func .= " mov QWORD${PTR}[8+rsp],rdi\t;WIN64 prologue\n"; 398 $func .= " mov QWORD${PTR}[16+rsp],rsi\n"; 399 $func .= " mov rax,rsp\n"; 400 $func .= "${decor}SEH_begin_$current_function->{name}:"; 401 $func .= ":" if ($masm); 402 $func .= "\n"; 403 my $narg = $current_function->{narg}; 404 $narg=6 if (!defined($narg)); 405 $func .= " mov rdi,rcx\n" if ($narg>0); 406 $func .= " mov rsi,rdx\n" if ($narg>1); 407 $func .= " mov rdx,r8\n" if ($narg>2); 408 $func .= " mov rcx,r9\n" if ($narg>3); 409 $func .= " mov r8,QWORD${PTR}[40+rsp]\n" if ($narg>4); 410 $func .= " mov r9,QWORD${PTR}[48+rsp]\n" if ($narg>5); 411 $func .= "\n"; 412 } else { 413 "$current_function->{name}". 414 ($nasm ? ":" : "\tPROC $current_function->{scope}"); 415 } 416 } 417} 418{ package expr; # pick up expressions 419 sub re { 420 my $self = shift; # single instance is enough... 421 local *line = shift; 422 undef $ret; 423 424 if ($line =~ /(^[^,]+)/) { 425 $self->{value} = $1; 426 $ret = $self; 427 $line = substr($line,@+[0]); $line =~ s/^\s+//; 428 429 $self->{value} =~ s/\@PLT// if (!$elf); 430 $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 431 $self->{value} =~ s/\.L/$decor/g; 432 } 433 $ret; 434 } 435 sub out { 436 my $self = shift; 437 if ($nasm && opcode->mnemonic()=~m/^j(?![re]cxz)/) { 438 "NEAR ".$self->{value}; 439 } else { 440 $self->{value}; 441 } 442 } 443} 444{ package directive; # pick up directives, which start with . 445 sub re { 446 my $self = shift; # single instance is enough... 447 local *line = shift; 448 undef $ret; 449 my $dir; 450 my %opcode = # lea 2f-1f(%rip),%dst; 1: nop; 2: 451 ( "%rax"=>0x01058d48, "%rcx"=>0x010d8d48, 452 "%rdx"=>0x01158d48, "%rbx"=>0x011d8d48, 453 "%rsp"=>0x01258d48, "%rbp"=>0x012d8d48, 454 "%rsi"=>0x01358d48, "%rdi"=>0x013d8d48, 455 "%r8" =>0x01058d4c, "%r9" =>0x010d8d4c, 456 "%r10"=>0x01158d4c, "%r11"=>0x011d8d4c, 457 "%r12"=>0x01258d4c, "%r13"=>0x012d8d4c, 458 "%r14"=>0x01358d4c, "%r15"=>0x013d8d4c ); 459 460 if ($line =~ /^\s*(\.\w+)/) { 461 $dir = $1; 462 $ret = $self; 463 undef $self->{value}; 464 $line = substr($line,@+[0]); $line =~ s/^\s+//; 465 466 SWITCH: for ($dir) { 467 /\.picmeup/ && do { if ($line =~ /(%r[\w]+)/i) { 468 $dir="\t.long"; 469 $line=sprintf "0x%x,0x90000000",$opcode{$1}; 470 } 471 last; 472 }; 473 /\.global|\.globl|\.extern/ 474 && do { $globals{$line} = $prefix . $line; 475 $line = $globals{$line} if ($prefix); 476 last; 477 }; 478 /\.type/ && do { ($sym,$type,$narg) = split(',',$line); 479 if ($type eq "\@function") { 480 undef $current_function; 481 $current_function->{name} = $sym; 482 $current_function->{abi} = "svr4"; 483 $current_function->{narg} = $narg; 484 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 485 } elsif ($type eq "\@abi-omnipotent") { 486 undef $current_function; 487 $current_function->{name} = $sym; 488 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 489 } 490 $line =~ s/\@abi\-omnipotent/\@function/; 491 $line =~ s/\@function.*/\@function/; 492 last; 493 }; 494 /\.asciz/ && do { if ($line =~ /^"(.*)"$/) { 495 $dir = ".byte"; 496 $line = join(",",unpack("C*",$1),0); 497 } 498 last; 499 }; 500 /\.rva|\.long|\.quad/ 501 && do { $line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 502 $line =~ s/\.L/$decor/g; 503 last; 504 }; 505 } 506 507 if ($gas) { 508 $self->{value} = $dir . "\t" . $line; 509 510 if ($dir =~ /\.extern/) { 511 $self->{value} = ""; # swallow extern 512 } elsif (!$elf && $dir =~ /\.type/) { 513 $self->{value} = ""; 514 $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" . 515 (defined($globals{$1})?".scl 2;":".scl 3;") . 516 "\t.type 32;\t.endef" 517 if ($win64 && $line =~ /([^,]+),\@function/); 518 } elsif (!$elf && $dir =~ /\.size/) { 519 $self->{value} = ""; 520 if (defined($current_function)) { 521 $self->{value} .= "${decor}SEH_end_$current_function->{name}:" 522 if ($win64 && $current_function->{abi} eq "svr4"); 523 undef $current_function; 524 } 525 } elsif (!$elf && $dir =~ /\.align/) { 526 $self->{value} = ".p2align\t" . (log($line)/log(2)); 527 } elsif ($dir eq ".section") { 528 $current_segment=$line; 529 if (!$elf && $current_segment eq ".init") { 530 if ($flavour eq "macosx") { $self->{value} = ".mod_init_func"; } 531 elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.ctors"; } 532 } 533 } elsif ($dir =~ /\.(text|data)/) { 534 $current_segment=".$1"; 535 } elsif ($dir =~ /\.hidden/) { 536 if ($flavour eq "macosx") { $self->{value} = ".private_extern\t$prefix$line"; } 537 elsif ($flavour eq "mingw64") { $self->{value} = ""; } 538 } elsif ($dir =~ /\.comm/) { 539 $self->{value} = "$dir\t$prefix$line"; 540 $self->{value} =~ s|,([0-9]+),([0-9]+)$|",$1,".log($2)/log(2)|e if ($flavour eq "macosx"); 541 } 542 $line = ""; 543 return $self; 544 } 545 546 # non-gas case or nasm/masm 547 SWITCH: for ($dir) { 548 /\.text/ && do { my $v=undef; 549 if ($nasm) { 550 $v="section .text code align=64\n"; 551 } else { 552 $v="$current_segment\tENDS\n" if ($current_segment); 553 $current_segment = ".text\$"; 554 $v.="$current_segment\tSEGMENT "; 555 $v.=$masm>=$masmref ? "ALIGN(256)" : "PAGE"; 556 $v.=" 'CODE'"; 557 } 558 $self->{value} = $v; 559 last; 560 }; 561 /\.data/ && do { my $v=undef; 562 if ($nasm) { 563 $v="section .data data align=8\n"; 564 } else { 565 $v="$current_segment\tENDS\n" if ($current_segment); 566 $current_segment = "_DATA"; 567 $v.="$current_segment\tSEGMENT"; 568 } 569 $self->{value} = $v; 570 last; 571 }; 572 /\.section/ && do { my $v=undef; 573 $line =~ s/([^,]*).*/$1/; 574 $line = ".CRT\$XCU" if ($line eq ".init"); 575 if ($nasm) { 576 $v="section $line"; 577 if ($line=~/\.([px])data/) { 578 $v.=" rdata align="; 579 $v.=$1 eq "p"? 4 : 8; 580 } elsif ($line=~/\.CRT\$/i) { 581 $v.=" rdata align=8"; 582 } 583 } else { 584 $v="$current_segment\tENDS\n" if ($current_segment); 585 $v.="$line\tSEGMENT"; 586 if ($line=~/\.([px])data/) { 587 $v.=" READONLY"; 588 $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref); 589 } elsif ($line=~/\.CRT\$/i) { 590 $v.=" READONLY "; 591 $v.=$masm>=$masmref ? "ALIGN(8)" : "DWORD"; 592 } 593 } 594 $current_segment = $line; 595 $self->{value} = $v; 596 last; 597 }; 598 /\.extern/ && do { $self->{value} = "EXTERN\t".$line; 599 $self->{value} .= ":NEAR" if ($masm); 600 last; 601 }; 602 /\.globl|.global/ 603 && do { $self->{value} = $masm?"PUBLIC":"global"; 604 $self->{value} .= "\t".$line; 605 last; 606 }; 607 /\.size/ && do { if (defined($current_function)) { 608 undef $self->{value}; 609 if ($current_function->{abi} eq "svr4") { 610 $self->{value}="${decor}SEH_end_$current_function->{name}:"; 611 $self->{value}.=":\n" if($masm); 612 } 613 $self->{value}.="$current_function->{name}\tENDP" if($masm && $current_function->{name}); 614 undef $current_function; 615 } 616 last; 617 }; 618 /\.align/ && do { $self->{value} = "ALIGN\t".$line; last; }; 619 /\.(value|long|rva|quad)/ 620 && do { my $sz = substr($1,0,1); 621 my @arr = split(/,\s*/,$line); 622 my $last = pop(@arr); 623 my $conv = sub { my $var=shift; 624 $var=~s/^(0b[0-1]+)/oct($1)/eig; 625 $var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm); 626 if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva")) 627 { $var=~s/([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; } 628 $var; 629 }; 630 631 $sz =~ tr/bvlrq/BWDDQ/; 632 $self->{value} = "\tD$sz\t"; 633 for (@arr) { $self->{value} .= &$conv($_).","; } 634 $self->{value} .= &$conv($last); 635 last; 636 }; 637 /\.byte/ && do { my @str=split(/,\s*/,$line); 638 map(s/(0b[0-1]+)/oct($1)/eig,@str); 639 map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm); 640 while ($#str>15) { 641 $self->{value}.="DB\t" 642 .join(",",@str[0..15])."\n"; 643 foreach (0..15) { shift @str; } 644 } 645 $self->{value}.="DB\t" 646 .join(",",@str) if (@str); 647 last; 648 }; 649 /\.comm/ && do { my @str=split(/,\s*/,$line); 650 my $v=undef; 651 if ($nasm) { 652 $v.="common $prefix@str[0] @str[1]"; 653 } else { 654 $v="$current_segment\tENDS\n" if ($current_segment); 655 $current_segment = "_DATA"; 656 $v.="$current_segment\tSEGMENT\n"; 657 $v.="COMM @str[0]:DWORD:".@str[1]/4; 658 } 659 $self->{value} = $v; 660 last; 661 }; 662 } 663 $line = ""; 664 } 665 666 $ret; 667 } 668 sub out { 669 my $self = shift; 670 $self->{value}; 671 } 672} 673 674sub rex { 675 local *opcode=shift; 676 my ($dst,$src,$rex)=@_; 677 678 $rex|=0x04 if($dst>=8); 679 $rex|=0x01 if($src>=8); 680 push @opcode,($rex|0x40) if ($rex); 681} 682 683# older gas and ml64 don't handle SSE>2 instructions 684my %regrm = ( "%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3, 685 "%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7 ); 686 687my $movq = sub { # elderly gas can't handle inter-register movq 688 my $arg = shift; 689 my @opcode=(0x66); 690 if ($arg =~ /%xmm([0-9]+),\s*%r(\w+)/) { 691 my ($src,$dst)=($1,$2); 692 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 693 rex(\@opcode,$src,$dst,0x8); 694 push @opcode,0x0f,0x7e; 695 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 696 @opcode; 697 } elsif ($arg =~ /%r(\w+),\s*%xmm([0-9]+)/) { 698 my ($src,$dst)=($2,$1); 699 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 700 rex(\@opcode,$src,$dst,0x8); 701 push @opcode,0x0f,0x6e; 702 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 703 @opcode; 704 } else { 705 (); 706 } 707}; 708 709my $pextrd = sub { 710 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*(%\w+)/) { 711 my @opcode=(0x66); 712 $imm=$1; 713 $src=$2; 714 $dst=$3; 715 if ($dst =~ /%r([0-9]+)d/) { $dst = $1; } 716 elsif ($dst =~ /%e/) { $dst = $regrm{$dst}; } 717 rex(\@opcode,$src,$dst); 718 push @opcode,0x0f,0x3a,0x16; 719 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 720 push @opcode,$imm; 721 @opcode; 722 } else { 723 (); 724 } 725}; 726 727my $pinsrd = sub { 728 if (shift =~ /\$([0-9]+),\s*(%\w+),\s*%xmm([0-9]+)/) { 729 my @opcode=(0x66); 730 $imm=$1; 731 $src=$2; 732 $dst=$3; 733 if ($src =~ /%r([0-9]+)/) { $src = $1; } 734 elsif ($src =~ /%e/) { $src = $regrm{$src}; } 735 rex(\@opcode,$dst,$src); 736 push @opcode,0x0f,0x3a,0x22; 737 push @opcode,0xc0|(($dst&7)<<3)|($src&7); # ModR/M 738 push @opcode,$imm; 739 @opcode; 740 } else { 741 (); 742 } 743}; 744 745my $pshufb = sub { 746 if (shift =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) { 747 my @opcode=(0x66); 748 rex(\@opcode,$2,$1); 749 push @opcode,0x0f,0x38,0x00; 750 push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M 751 @opcode; 752 } else { 753 (); 754 } 755}; 756 757my $palignr = sub { 758 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 759 my @opcode=(0x66); 760 rex(\@opcode,$3,$2); 761 push @opcode,0x0f,0x3a,0x0f; 762 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 763 push @opcode,$1; 764 @opcode; 765 } else { 766 (); 767 } 768}; 769 770my $pclmulqdq = sub { 771 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 772 my @opcode=(0x66); 773 rex(\@opcode,$3,$2); 774 push @opcode,0x0f,0x3a,0x44; 775 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 776 my $c=$1; 777 push @opcode,$c=~/^0/?oct($c):$c; 778 @opcode; 779 } else { 780 (); 781 } 782}; 783 784my $rdrand = sub { 785 if (shift =~ /%[er](\w+)/) { 786 my @opcode=(); 787 my $dst=$1; 788 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 789 rex(\@opcode,0,$1,8); 790 push @opcode,0x0f,0xc7,0xf0|($dst&7); 791 @opcode; 792 } else { 793 (); 794 } 795}; 796 797my $rdseed = sub { 798 if (shift =~ /%[er](\w+)/) { 799 my @opcode=(); 800 my $dst=$1; 801 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 802 rex(\@opcode,0,$1,8); 803 push @opcode,0x0f,0xc7,0xf8|($dst&7); 804 @opcode; 805 } else { 806 (); 807 } 808}; 809 810sub rxb { 811 local *opcode=shift; 812 my ($dst,$src1,$src2,$rxb)=@_; 813 814 $rxb|=0x7<<5; 815 $rxb&=~(0x04<<5) if($dst>=8); 816 $rxb&=~(0x01<<5) if($src1>=8); 817 $rxb&=~(0x02<<5) if($src2>=8); 818 push @opcode,$rxb; 819} 820 821my $vprotd = sub { 822 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 823 my @opcode=(0x8f); 824 rxb(\@opcode,$3,$2,-1,0x08); 825 push @opcode,0x78,0xc2; 826 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 827 my $c=$1; 828 push @opcode,$c=~/^0/?oct($c):$c; 829 @opcode; 830 } else { 831 (); 832 } 833}; 834 835my $vprotq = sub { 836 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 837 my @opcode=(0x8f); 838 rxb(\@opcode,$3,$2,-1,0x08); 839 push @opcode,0x78,0xc3; 840 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 841 my $c=$1; 842 push @opcode,$c=~/^0/?oct($c):$c; 843 @opcode; 844 } else { 845 (); 846 } 847}; 848 849if ($nasm) { 850 print <<___; 851default rel 852%define XMMWORD 853%define YMMWORD 854%define ZMMWORD 855___ 856} elsif ($masm) { 857 print <<___; 858OPTION DOTNAME 859___ 860} 861while($line=<>) { 862 863 chomp($line); 864 865 $line =~ s|[#!].*$||; # get rid of asm-style comments... 866 $line =~ s|/\*.*\*/||; # ... and C-style comments... 867 $line =~ s|^\s+||; # ... and skip white spaces in beginning 868 $line =~ s|\s+$||; # ... and at the end 869 870 undef $label; 871 undef $opcode; 872 undef @args; 873 874 if ($label=label->re(\$line)) { print $label->out(); } 875 876 if (directive->re(\$line)) { 877 printf "%s",directive->out(); 878 } elsif ($opcode=opcode->re(\$line)) { 879 my $asm = eval("\$".$opcode->mnemonic()); 880 undef @bytes; 881 882 if ((ref($asm) eq 'CODE') && scalar(@bytes=&$asm($line))) { 883 print $gas?".byte\t":"DB\t",join(',',@bytes),"\n"; 884 next; 885 } 886 887 ARGUMENT: while (1) { 888 my $arg; 889 890 if ($arg=register->re(\$line)) { opcode->size($arg->size()); } 891 elsif ($arg=const->re(\$line)) { } 892 elsif ($arg=ea->re(\$line)) { } 893 elsif ($arg=expr->re(\$line)) { } 894 else { last ARGUMENT; } 895 896 push @args,$arg; 897 898 last ARGUMENT if ($line !~ /^,/); 899 900 $line =~ s/^,\s*//; 901 } # ARGUMENT: 902 903 if ($#args>=0) { 904 my $insn; 905 my $sz=opcode->size(); 906 907 if ($gas) { 908 $insn = $opcode->out($#args>=1?$args[$#args]->size():$sz); 909 @args = map($_->out($sz),@args); 910 printf "\t%s\t%s",$insn,join(",",@args); 911 } else { 912 $insn = $opcode->out(); 913 foreach (@args) { 914 my $arg = $_->out(); 915 # $insn.=$sz compensates for movq, pinsrw, ... 916 if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; } 917 if ($arg =~ /^ymm[0-9]+$/) { $insn.=$sz; $sz="y" if(!$sz); last; } 918 if ($arg =~ /^zmm[0-9]+$/) { $insn.=$sz; $sz="z" if(!$sz); last; } 919 if ($arg =~ /^mm[0-9]+$/) { $insn.=$sz; $sz="q" if(!$sz); last; } 920 } 921 @args = reverse(@args); 922 undef $sz if ($nasm && $opcode->mnemonic() eq "lea"); 923 printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args)); 924 } 925 } else { 926 printf "\t%s",$opcode->out(); 927 } 928 } 929 930 print $line,"\n"; 931} 932 933print "\n$current_segment\tENDS\n" if ($current_segment && $masm); 934print "END\n" if ($masm); 935 936close STDOUT; 937 938 ################################################# 939# Cross-reference x86_64 ABI "card" 940# 941# Unix Win64 942# %rax * * 943# %rbx - - 944# %rcx #4 #1 945# %rdx #3 #2 946# %rsi #2 - 947# %rdi #1 - 948# %rbp - - 949# %rsp - - 950# %r8 #5 #3 951# %r9 #6 #4 952# %r10 * * 953# %r11 * * 954# %r12 - - 955# %r13 - - 956# %r14 - - 957# %r15 - - 958# 959# (*) volatile register 960# (-) preserved by callee 961# (#) Nth argument, volatile 962# 963# In Unix terms top of stack is argument transfer area for arguments 964# which could not be accomodated in registers. Or in other words 7th 965# [integer] argument resides at 8(%rsp) upon function entry point. 966# 128 bytes above %rsp constitute a "red zone" which is not touched 967# by signal handlers and can be used as temporal storage without 968# allocating a frame. 969# 970# In Win64 terms N*8 bytes on top of stack is argument transfer area, 971# which belongs to/can be overwritten by callee. N is the number of 972# arguments passed to callee, *but* not less than 4! This means that 973# upon function entry point 5th argument resides at 40(%rsp), as well 974# as that 32 bytes from 8(%rsp) can always be used as temporal 975# storage [without allocating a frame]. One can actually argue that 976# one can assume a "red zone" above stack pointer under Win64 as well. 977# Point is that at apparently no occasion Windows kernel would alter 978# the area above user stack pointer in true asynchronous manner... 979# 980# All the above means that if assembler programmer adheres to Unix 981# register and stack layout, but disregards the "red zone" existence, 982# it's possible to use following prologue and epilogue to "gear" from 983# Unix to Win64 ABI in leaf functions with not more than 6 arguments. 984# 985# omnipotent_function: 986# ifdef WIN64 987# movq %rdi,8(%rsp) 988# movq %rsi,16(%rsp) 989# movq %rcx,%rdi ; if 1st argument is actually present 990# movq %rdx,%rsi ; if 2nd argument is actually ... 991# movq %r8,%rdx ; if 3rd argument is ... 992# movq %r9,%rcx ; if 4th argument ... 993# movq 40(%rsp),%r8 ; if 5th ... 994# movq 48(%rsp),%r9 ; if 6th ... 995# endif 996# ... 997# ifdef WIN64 998# movq 8(%rsp),%rdi 999# movq 16(%rsp),%rsi 1000# endif 1001# ret 1002# 1003 ################################################# 1004# Win64 SEH, Structured Exception Handling. 1005# 1006# Unlike on Unix systems(*) lack of Win64 stack unwinding information 1007# has undesired side-effect at run-time: if an exception is raised in 1008# assembler subroutine such as those in question (basically we're 1009# referring to segmentation violations caused by malformed input 1010# parameters), the application is briskly terminated without invoking 1011# any exception handlers, most notably without generating memory dump 1012# or any user notification whatsoever. This poses a problem. It's 1013# possible to address it by registering custom language-specific 1014# handler that would restore processor context to the state at 1015# subroutine entry point and return "exception is not handled, keep 1016# unwinding" code. Writing such handler can be a challenge... But it's 1017# doable, though requires certain coding convention. Consider following 1018# snippet: 1019# 1020# .type function,@function 1021# function: 1022# movq %rsp,%rax # copy rsp to volatile register 1023# pushq %r15 # save non-volatile registers 1024# pushq %rbx 1025# pushq %rbp 1026# movq %rsp,%r11 1027# subq %rdi,%r11 # prepare [variable] stack frame 1028# andq $-64,%r11 1029# movq %rax,0(%r11) # check for exceptions 1030# movq %r11,%rsp # allocate [variable] stack frame 1031# movq %rax,0(%rsp) # save original rsp value 1032# magic_point: 1033# ... 1034# movq 0(%rsp),%rcx # pull original rsp value 1035# movq -24(%rcx),%rbp # restore non-volatile registers 1036# movq -16(%rcx),%rbx 1037# movq -8(%rcx),%r15 1038# movq %rcx,%rsp # restore original rsp 1039# ret 1040# .size function,.-function 1041# 1042# The key is that up to magic_point copy of original rsp value remains 1043# in chosen volatile register and no non-volatile register, except for 1044# rsp, is modified. While past magic_point rsp remains constant till 1045# the very end of the function. In this case custom language-specific 1046# exception handler would look like this: 1047# 1048# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, 1049# CONTEXT *context,DISPATCHER_CONTEXT *disp) 1050# { ULONG64 *rsp = (ULONG64 *)context->Rax; 1051# if (context->Rip >= magic_point) 1052# { rsp = ((ULONG64 **)context->Rsp)[0]; 1053# context->Rbp = rsp[-3]; 1054# context->Rbx = rsp[-2]; 1055# context->R15 = rsp[-1]; 1056# } 1057# context->Rsp = (ULONG64)rsp; 1058# context->Rdi = rsp[1]; 1059# context->Rsi = rsp[2]; 1060# 1061# memcpy (disp->ContextRecord,context,sizeof(CONTEXT)); 1062# RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase, 1063# dips->ControlPc,disp->FunctionEntry,disp->ContextRecord, 1064# &disp->HandlerData,&disp->EstablisherFrame,NULL); 1065# return ExceptionContinueSearch; 1066# } 1067# 1068# It's appropriate to implement this handler in assembler, directly in 1069# function's module. In order to do that one has to know members' 1070# offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant 1071# values. Here they are: 1072# 1073# CONTEXT.Rax 120 1074# CONTEXT.Rcx 128 1075# CONTEXT.Rdx 136 1076# CONTEXT.Rbx 144 1077# CONTEXT.Rsp 152 1078# CONTEXT.Rbp 160 1079# CONTEXT.Rsi 168 1080# CONTEXT.Rdi 176 1081# CONTEXT.R8 184 1082# CONTEXT.R9 192 1083# CONTEXT.R10 200 1084# CONTEXT.R11 208 1085# CONTEXT.R12 216 1086# CONTEXT.R13 224 1087# CONTEXT.R14 232 1088# CONTEXT.R15 240 1089# CONTEXT.Rip 248 1090# CONTEXT.Xmm6 512 1091# sizeof(CONTEXT) 1232 1092# DISPATCHER_CONTEXT.ControlPc 0 1093# DISPATCHER_CONTEXT.ImageBase 8 1094# DISPATCHER_CONTEXT.FunctionEntry 16 1095# DISPATCHER_CONTEXT.EstablisherFrame 24 1096# DISPATCHER_CONTEXT.TargetIp 32 1097# DISPATCHER_CONTEXT.ContextRecord 40 1098# DISPATCHER_CONTEXT.LanguageHandler 48 1099# DISPATCHER_CONTEXT.HandlerData 56 1100# UNW_FLAG_NHANDLER 0 1101# ExceptionContinueSearch 1 1102# 1103# In order to tie the handler to the function one has to compose 1104# couple of structures: one for .xdata segment and one for .pdata. 1105# 1106# UNWIND_INFO structure for .xdata segment would be 1107# 1108# function_unwind_info: 1109# .byte 9,0,0,0 1110# .rva handler 1111# 1112# This structure designates exception handler for a function with 1113# zero-length prologue, no stack frame or frame register. 1114# 1115# To facilitate composing of .pdata structures, auto-generated "gear" 1116# prologue copies rsp value to rax and denotes next instruction with 1117# .LSEH_begin_{function_name} label. This essentially defines the SEH 1118# styling rule mentioned in the beginning. Position of this label is 1119# chosen in such manner that possible exceptions raised in the "gear" 1120# prologue would be accounted to caller and unwound from latter's frame. 1121# End of function is marked with respective .LSEH_end_{function_name} 1122# label. To summarize, .pdata segment would contain 1123# 1124# .rva .LSEH_begin_function 1125# .rva .LSEH_end_function 1126# .rva function_unwind_info 1127# 1128# Reference to functon_unwind_info from .xdata segment is the anchor. 1129# In case you wonder why references are 32-bit .rvas and not 64-bit 1130# .quads. References put into these two segments are required to be 1131# *relative* to the base address of the current binary module, a.k.a. 1132# image base. No Win64 module, be it .exe or .dll, can be larger than 1133# 2GB and thus such relative references can be and are accommodated in 1134# 32 bits. 1135# 1136# Having reviewed the example function code, one can argue that "movq 1137# %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix 1138# rax would contain an undefined value. If this "offends" you, use 1139# another register and refrain from modifying rax till magic_point is 1140# reached, i.e. as if it was a non-volatile register. If more registers 1141# are required prior [variable] frame setup is completed, note that 1142# nobody says that you can have only one "magic point." You can 1143# "liberate" non-volatile registers by denoting last stack off-load 1144# instruction and reflecting it in finer grade unwind logic in handler. 1145# After all, isn't it why it's called *language-specific* handler... 1146# 1147# Attentive reader can notice that exceptions would be mishandled in 1148# auto-generated "gear" epilogue. Well, exception effectively can't 1149# occur there, because if memory area used by it was subject to 1150# segmentation violation, then it would be raised upon call to the 1151# function (and as already mentioned be accounted to caller, which is 1152# not a problem). If you're still not comfortable, then define tail 1153# "magic point" just prior ret instruction and have handler treat it... 1154# 1155# (*) Note that we're talking about run-time, not debug-time. Lack of 1156# unwind information makes debugging hard on both Windows and 1157# Unix. "Unlike" referes to the fact that on Unix signal handler 1158# will always be invoked, core dumped and appropriate exit code 1159# returned to parent (for user notification). 1160