x86_64-xlate.pl revision 312826
1#!/usr/bin/env perl 2 3# Ascetic x86_64 AT&T to MASM/NASM assembler translator by <appro>. 4# 5# Why AT&T to MASM and not vice versa? Several reasons. Because AT&T 6# format is way easier to parse. Because it's simpler to "gear" from 7# Unix ABI to Windows one [see cross-reference "card" at the end of 8# file]. Because Linux targets were available first... 9# 10# In addition the script also "distills" code suitable for GNU 11# assembler, so that it can be compiled with more rigid assemblers, 12# such as Solaris /usr/ccs/bin/as. 13# 14# This translator is not designed to convert *arbitrary* assembler 15# code from AT&T format to MASM one. It's designed to convert just 16# enough to provide for dual-ABI OpenSSL modules development... 17# There *are* limitations and you might have to modify your assembler 18# code or this script to achieve the desired result... 19# 20# Currently recognized limitations: 21# 22# - can't use multiple ops per line; 23# 24# Dual-ABI styling rules. 25# 26# 1. Adhere to Unix register and stack layout [see cross-reference 27# ABI "card" at the end for explanation]. 28# 2. Forget about "red zone," stick to more traditional blended 29# stack frame allocation. If volatile storage is actually required 30# that is. If not, just leave the stack as is. 31# 3. Functions tagged with ".type name,@function" get crafted with 32# unified Win64 prologue and epilogue automatically. If you want 33# to take care of ABI differences yourself, tag functions as 34# ".type name,@abi-omnipotent" instead. 35# 4. To optimize the Win64 prologue you can specify number of input 36# arguments as ".type name,@function,N." Keep in mind that if N is 37# larger than 6, then you *have to* write "abi-omnipotent" code, 38# because >6 cases can't be addressed with unified prologue. 39# 5. Name local labels as .L*, do *not* use dynamic labels such as 1: 40# (sorry about latter). 41# 6. Don't use [or hand-code with .byte] "rep ret." "ret" mnemonic is 42# required to identify the spots, where to inject Win64 epilogue! 43# But on the pros, it's then prefixed with rep automatically:-) 44# 7. Stick to explicit ip-relative addressing. If you have to use 45# GOTPCREL addressing, stick to mov symbol@GOTPCREL(%rip),%r??. 46# Both are recognized and translated to proper Win64 addressing 47# modes. To support legacy code a synthetic directive, .picmeup, 48# is implemented. It puts address of the *next* instruction into 49# target register, e.g.: 50# 51# .picmeup %rax 52# lea .Label-.(%rax),%rax 53# 54# 8. In order to provide for structured exception handling unified 55# Win64 prologue copies %rsp value to %rax. For further details 56# see SEH paragraph at the end. 57# 9. .init segment is allowed to contain calls to functions only. 58# a. If function accepts more than 4 arguments *and* >4th argument 59# is declared as non 64-bit value, do clear its upper part. 60 61my $flavour = shift; 62my $output = shift; 63if ($flavour =~ /\./) { $output = $flavour; undef $flavour; } 64 65open STDOUT,">$output" || die "can't open $output: $!" 66 if (defined($output)); 67 68my $gas=1; $gas=0 if ($output =~ /\.asm$/); 69my $elf=1; $elf=0 if (!$gas); 70my $win64=0; 71my $prefix=""; 72my $decor=".L"; 73 74my $masmref=8 + 50727*2**-32; # 8.00.50727 shipped with VS2005 75my $masm=0; 76my $PTR=" PTR"; 77 78my $nasmref=2.03; 79my $nasm=0; 80 81if ($flavour eq "mingw64") { $gas=1; $elf=0; $win64=1; 82 $prefix=`echo __USER_LABEL_PREFIX__ | $ENV{CC} -E -P -`; 83 chomp($prefix); 84 } 85elsif ($flavour eq "macosx") { $gas=1; $elf=0; $prefix="_"; $decor="L\$"; } 86elsif ($flavour eq "masm") { $gas=0; $elf=0; $masm=$masmref; $win64=1; $decor="\$L\$"; } 87elsif ($flavour eq "nasm") { $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $decor="\$L\$"; $PTR=""; } 88elsif (!$gas) 89{ if ($ENV{ASM} =~ m/nasm/ && `nasm -v` =~ m/version ([0-9]+)\.([0-9]+)/i) 90 { $nasm = $1 + $2*0.01; $PTR=""; } 91 elsif (`ml64 2>&1` =~ m/Version ([0-9]+)\.([0-9]+)(\.([0-9]+))?/) 92 { $masm = $1 + $2*2**-16 + $4*2**-32; } 93 die "no assembler found on %PATH" if (!($nasm || $masm)); 94 $win64=1; 95 $elf=0; 96 $decor="\$L\$"; 97} 98 99my $current_segment; 100my $current_function; 101my %globals; 102 103{ package opcode; # pick up opcodes 104 sub re { 105 my $self = shift; # single instance in enough... 106 local *line = shift; 107 undef $ret; 108 109 if ($line =~ /^([a-z][a-z0-9]*)/i) { 110 $self->{op} = $1; 111 $ret = $self; 112 $line = substr($line,@+[0]); $line =~ s/^\s+//; 113 114 undef $self->{sz}; 115 if ($self->{op} =~ /^(movz)x?([bw]).*/) { # movz is pain... 116 $self->{op} = $1; 117 $self->{sz} = $2; 118 } elsif ($self->{op} =~ /call|jmp/) { 119 $self->{sz} = ""; 120 } elsif ($self->{op} =~ /^p/ && $' !~ /^(ush|op|insrw)/) { # SSEn 121 $self->{sz} = ""; 122 } elsif ($self->{op} =~ /^v/) { # VEX 123 $self->{sz} = ""; 124 } elsif ($self->{op} =~ /mov[dq]/ && $line =~ /%xmm/) { 125 $self->{sz} = ""; 126 } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) { 127 $self->{op} = $1; 128 $self->{sz} = $2; 129 } 130 } 131 $ret; 132 } 133 sub size { 134 my $self = shift; 135 my $sz = shift; 136 $self->{sz} = $sz if (defined($sz) && !defined($self->{sz})); 137 $self->{sz}; 138 } 139 sub out { 140 my $self = shift; 141 if ($gas) { 142 if ($self->{op} eq "movz") { # movz is pain... 143 sprintf "%s%s%s",$self->{op},$self->{sz},shift; 144 } elsif ($self->{op} =~ /^set/) { 145 "$self->{op}"; 146 } elsif ($self->{op} eq "ret") { 147 my $epilogue = ""; 148 if ($win64 && $current_function->{abi} eq "svr4") { 149 $epilogue = "movq 8(%rsp),%rdi\n\t" . 150 "movq 16(%rsp),%rsi\n\t"; 151 } 152 $epilogue . ".byte 0xf3,0xc3"; 153 } elsif ($self->{op} eq "call" && !$elf && $current_segment eq ".init") { 154 ".p2align\t3\n\t.quad"; 155 } else { 156 "$self->{op}$self->{sz}"; 157 } 158 } else { 159 $self->{op} =~ s/^movz/movzx/; 160 if ($self->{op} eq "ret") { 161 $self->{op} = ""; 162 if ($win64 && $current_function->{abi} eq "svr4") { 163 $self->{op} = "mov rdi,QWORD${PTR}[8+rsp]\t;WIN64 epilogue\n\t". 164 "mov rsi,QWORD${PTR}[16+rsp]\n\t"; 165 } 166 $self->{op} .= "DB\t0F3h,0C3h\t\t;repret"; 167 } elsif ($self->{op} =~ /^(pop|push)f/) { 168 $self->{op} .= $self->{sz}; 169 } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") { 170 $self->{op} = "\tDQ"; 171 } 172 $self->{op}; 173 } 174 } 175 sub mnemonic { 176 my $self=shift; 177 my $op=shift; 178 $self->{op}=$op if (defined($op)); 179 $self->{op}; 180 } 181} 182{ package const; # pick up constants, which start with $ 183 sub re { 184 my $self = shift; # single instance in enough... 185 local *line = shift; 186 undef $ret; 187 188 if ($line =~ /^\$([^,]+)/) { 189 $self->{value} = $1; 190 $ret = $self; 191 $line = substr($line,@+[0]); $line =~ s/^\s+//; 192 } 193 $ret; 194 } 195 sub out { 196 my $self = shift; 197 198 $self->{value} =~ s/\b(0b[0-1]+)/oct($1)/eig; 199 if ($gas) { 200 # Solaris /usr/ccs/bin/as can't handle multiplications 201 # in $self->{value} 202 my $value = $self->{value}; 203 $value =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi; 204 if ($value =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg) { 205 $self->{value} = $value; 206 } 207 sprintf "\$%s",$self->{value}; 208 } else { 209 $self->{value} =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm); 210 sprintf "%s",$self->{value}; 211 } 212 } 213} 214{ package ea; # pick up effective addresses: expr(%reg,%reg,scale) 215 sub re { 216 my $self = shift; # single instance in enough... 217 local *line = shift; 218 undef $ret; 219 220 # optional * ---vvv--- appears in indirect jmp/call 221 if ($line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)/) { 222 $self->{asterisk} = $1; 223 $self->{label} = $2; 224 ($self->{base},$self->{index},$self->{scale})=split(/,/,$3); 225 $self->{scale} = 1 if (!defined($self->{scale})); 226 $ret = $self; 227 $line = substr($line,@+[0]); $line =~ s/^\s+//; 228 229 if ($win64 && $self->{label} =~ s/\@GOTPCREL//) { 230 die if (opcode->mnemonic() ne "mov"); 231 opcode->mnemonic("lea"); 232 } 233 $self->{base} =~ s/^%//; 234 $self->{index} =~ s/^%// if (defined($self->{index})); 235 } 236 $ret; 237 } 238 sub size {} 239 sub out { 240 my $self = shift; 241 my $sz = shift; 242 243 $self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 244 $self->{label} =~ s/\.L/$decor/g; 245 246 # Silently convert all EAs to 64-bit. This is required for 247 # elder GNU assembler and results in more compact code, 248 # *but* most importantly AES module depends on this feature! 249 $self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; 250 $self->{base} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; 251 252 # Solaris /usr/ccs/bin/as can't handle multiplications 253 # in $self->{label}... 254 use integer; 255 $self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi; 256 $self->{label} =~ s/\b([0-9]+\s*[\*\/\%]\s*[0-9]+)\b/eval($1)/eg; 257 258 # Some assemblers insist on signed presentation of 32-bit 259 # offsets, but sign extension is a tricky business in perl... 260 if ((1<<31)<<1) { 261 $self->{label} =~ s/\b([0-9]+)\b/$1<<32>>32/eg; 262 } else { 263 $self->{label} =~ s/\b([0-9]+)\b/$1>>0/eg; 264 } 265 266 if (!$self->{label} && $self->{index} && $self->{scale}==1 && 267 $self->{base} =~ /(rbp|r13)/) { 268 $self->{base} = $self->{index}; $self->{index} = $1; 269 } 270 271 if ($gas) { 272 $self->{label} =~ s/^___imp_/__imp__/ if ($flavour eq "mingw64"); 273 274 if (defined($self->{index})) { 275 sprintf "%s%s(%s,%%%s,%d)",$self->{asterisk}, 276 $self->{label}, 277 $self->{base}?"%$self->{base}":"", 278 $self->{index},$self->{scale}; 279 } else { 280 sprintf "%s%s(%%%s)", $self->{asterisk},$self->{label},$self->{base}; 281 } 282 } else { 283 %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR", 284 l=>"DWORD$PTR", d=>"DWORD$PTR", 285 q=>"QWORD$PTR", o=>"OWORD$PTR", 286 x=>"XMMWORD$PTR", y=>"YMMWORD$PTR", z=>"ZMMWORD$PTR" ); 287 288 $self->{label} =~ s/\./\$/g; 289 $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig; 290 $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/); 291 292 ($self->{asterisk}) && ($sz="q") || 293 (opcode->mnemonic() =~ /^v?mov([qd])$/) && ($sz=$1) || 294 (opcode->mnemonic() =~ /^v?pinsr([qdwb])$/) && ($sz=$1) || 295 (opcode->mnemonic() =~ /^vpbroadcast([qdwb])$/) && ($sz=$1) || 296 (opcode->mnemonic() =~ /^vinsert[fi]128$/) && ($sz="x"); 297 298 if (defined($self->{index})) { 299 sprintf "%s[%s%s*%d%s]",$szmap{$sz}, 300 $self->{label}?"$self->{label}+":"", 301 $self->{index},$self->{scale}, 302 $self->{base}?"+$self->{base}":""; 303 } elsif ($self->{base} eq "rip") { 304 sprintf "%s[%s]",$szmap{$sz},$self->{label}; 305 } else { 306 sprintf "%s[%s%s]",$szmap{$sz}, 307 $self->{label}?"$self->{label}+":"", 308 $self->{base}; 309 } 310 } 311 } 312} 313{ package register; # pick up registers, which start with %. 314 sub re { 315 my $class = shift; # muliple instances... 316 my $self = {}; 317 local *line = shift; 318 undef $ret; 319 320 # optional * ---vvv--- appears in indirect jmp/call 321 if ($line =~ /^(\*?)%(\w+)/) { 322 bless $self,$class; 323 $self->{asterisk} = $1; 324 $self->{value} = $2; 325 $ret = $self; 326 $line = substr($line,@+[0]); $line =~ s/^\s+//; 327 } 328 $ret; 329 } 330 sub size { 331 my $self = shift; 332 undef $ret; 333 334 if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; } 335 elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; } 336 elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; } 337 elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; } 338 elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; } 339 elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; } 340 elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; } 341 elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; } 342 343 $ret; 344 } 345 sub out { 346 my $self = shift; 347 if ($gas) { sprintf "%s%%%s",$self->{asterisk},$self->{value}; } 348 else { $self->{value}; } 349 } 350} 351{ package label; # pick up labels, which end with : 352 sub re { 353 my $self = shift; # single instance is enough... 354 local *line = shift; 355 undef $ret; 356 357 if ($line =~ /(^[\.\w]+)\:/) { 358 $self->{value} = $1; 359 $ret = $self; 360 $line = substr($line,@+[0]); $line =~ s/^\s+//; 361 362 $self->{value} =~ s/^\.L/$decor/; 363 } 364 $ret; 365 } 366 sub out { 367 my $self = shift; 368 369 if ($gas) { 370 my $func = ($globals{$self->{value}} or $self->{value}) . ":"; 371 if ($win64 && 372 $current_function->{name} eq $self->{value} && 373 $current_function->{abi} eq "svr4") { 374 $func .= "\n"; 375 $func .= " movq %rdi,8(%rsp)\n"; 376 $func .= " movq %rsi,16(%rsp)\n"; 377 $func .= " movq %rsp,%rax\n"; 378 $func .= "${decor}SEH_begin_$current_function->{name}:\n"; 379 my $narg = $current_function->{narg}; 380 $narg=6 if (!defined($narg)); 381 $func .= " movq %rcx,%rdi\n" if ($narg>0); 382 $func .= " movq %rdx,%rsi\n" if ($narg>1); 383 $func .= " movq %r8,%rdx\n" if ($narg>2); 384 $func .= " movq %r9,%rcx\n" if ($narg>3); 385 $func .= " movq 40(%rsp),%r8\n" if ($narg>4); 386 $func .= " movq 48(%rsp),%r9\n" if ($narg>5); 387 } 388 $func; 389 } elsif ($self->{value} ne "$current_function->{name}") { 390 $self->{value} .= ":" if ($masm && $ret!~m/^\$/); 391 $self->{value} . ":"; 392 } elsif ($win64 && $current_function->{abi} eq "svr4") { 393 my $func = "$current_function->{name}" . 394 ($nasm ? ":" : "\tPROC $current_function->{scope}") . 395 "\n"; 396 $func .= " mov QWORD${PTR}[8+rsp],rdi\t;WIN64 prologue\n"; 397 $func .= " mov QWORD${PTR}[16+rsp],rsi\n"; 398 $func .= " mov rax,rsp\n"; 399 $func .= "${decor}SEH_begin_$current_function->{name}:"; 400 $func .= ":" if ($masm); 401 $func .= "\n"; 402 my $narg = $current_function->{narg}; 403 $narg=6 if (!defined($narg)); 404 $func .= " mov rdi,rcx\n" if ($narg>0); 405 $func .= " mov rsi,rdx\n" if ($narg>1); 406 $func .= " mov rdx,r8\n" if ($narg>2); 407 $func .= " mov rcx,r9\n" if ($narg>3); 408 $func .= " mov r8,QWORD${PTR}[40+rsp]\n" if ($narg>4); 409 $func .= " mov r9,QWORD${PTR}[48+rsp]\n" if ($narg>5); 410 $func .= "\n"; 411 } else { 412 "$current_function->{name}". 413 ($nasm ? ":" : "\tPROC $current_function->{scope}"); 414 } 415 } 416} 417{ package expr; # pick up expressioins 418 sub re { 419 my $self = shift; # single instance is enough... 420 local *line = shift; 421 undef $ret; 422 423 if ($line =~ /(^[^,]+)/) { 424 $self->{value} = $1; 425 $ret = $self; 426 $line = substr($line,@+[0]); $line =~ s/^\s+//; 427 428 $self->{value} =~ s/\@PLT// if (!$elf); 429 $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 430 $self->{value} =~ s/\.L/$decor/g; 431 } 432 $ret; 433 } 434 sub out { 435 my $self = shift; 436 if ($nasm && opcode->mnemonic()=~m/^j(?![re]cxz)/) { 437 "NEAR ".$self->{value}; 438 } else { 439 $self->{value}; 440 } 441 } 442} 443{ package directive; # pick up directives, which start with . 444 sub re { 445 my $self = shift; # single instance is enough... 446 local *line = shift; 447 undef $ret; 448 my $dir; 449 my %opcode = # lea 2f-1f(%rip),%dst; 1: nop; 2: 450 ( "%rax"=>0x01058d48, "%rcx"=>0x010d8d48, 451 "%rdx"=>0x01158d48, "%rbx"=>0x011d8d48, 452 "%rsp"=>0x01258d48, "%rbp"=>0x012d8d48, 453 "%rsi"=>0x01358d48, "%rdi"=>0x013d8d48, 454 "%r8" =>0x01058d4c, "%r9" =>0x010d8d4c, 455 "%r10"=>0x01158d4c, "%r11"=>0x011d8d4c, 456 "%r12"=>0x01258d4c, "%r13"=>0x012d8d4c, 457 "%r14"=>0x01358d4c, "%r15"=>0x013d8d4c ); 458 459 if ($line =~ /^\s*(\.\w+)/) { 460 $dir = $1; 461 $ret = $self; 462 undef $self->{value}; 463 $line = substr($line,@+[0]); $line =~ s/^\s+//; 464 465 SWITCH: for ($dir) { 466 /\.picmeup/ && do { if ($line =~ /(%r[\w]+)/i) { 467 $dir="\t.long"; 468 $line=sprintf "0x%x,0x90000000",$opcode{$1}; 469 } 470 last; 471 }; 472 /\.global|\.globl|\.extern/ 473 && do { $globals{$line} = $prefix . $line; 474 $line = $globals{$line} if ($prefix); 475 last; 476 }; 477 /\.type/ && do { ($sym,$type,$narg) = split(',',$line); 478 if ($type eq "\@function") { 479 undef $current_function; 480 $current_function->{name} = $sym; 481 $current_function->{abi} = "svr4"; 482 $current_function->{narg} = $narg; 483 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 484 } elsif ($type eq "\@abi-omnipotent") { 485 undef $current_function; 486 $current_function->{name} = $sym; 487 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 488 } 489 $line =~ s/\@abi\-omnipotent/\@function/; 490 $line =~ s/\@function.*/\@function/; 491 last; 492 }; 493 /\.asciz/ && do { if ($line =~ /^"(.*)"$/) { 494 $dir = ".byte"; 495 $line = join(",",unpack("C*",$1),0); 496 } 497 last; 498 }; 499 /\.rva|\.long|\.quad/ 500 && do { $line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 501 $line =~ s/\.L/$decor/g; 502 last; 503 }; 504 } 505 506 if ($gas) { 507 $self->{value} = $dir . "\t" . $line; 508 509 if ($dir =~ /\.extern/) { 510 $self->{value} = ""; # swallow extern 511 } elsif (!$elf && $dir =~ /\.type/) { 512 $self->{value} = ""; 513 $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" . 514 (defined($globals{$1})?".scl 2;":".scl 3;") . 515 "\t.type 32;\t.endef" 516 if ($win64 && $line =~ /([^,]+),\@function/); 517 } elsif (!$elf && $dir =~ /\.size/) { 518 $self->{value} = ""; 519 if (defined($current_function)) { 520 $self->{value} .= "${decor}SEH_end_$current_function->{name}:" 521 if ($win64 && $current_function->{abi} eq "svr4"); 522 undef $current_function; 523 } 524 } elsif (!$elf && $dir =~ /\.align/) { 525 $self->{value} = ".p2align\t" . (log($line)/log(2)); 526 } elsif ($dir eq ".section") { 527 $current_segment=$line; 528 if (!$elf && $current_segment eq ".init") { 529 if ($flavour eq "macosx") { $self->{value} = ".mod_init_func"; } 530 elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.ctors"; } 531 } 532 } elsif ($dir =~ /\.(text|data)/) { 533 $current_segment=".$1"; 534 } elsif ($dir =~ /\.hidden/) { 535 if ($flavour eq "macosx") { $self->{value} = ".private_extern\t$prefix$line"; } 536 elsif ($flavour eq "mingw64") { $self->{value} = ""; } 537 } elsif ($dir =~ /\.comm/) { 538 $self->{value} = "$dir\t$prefix$line"; 539 $self->{value} =~ s|,([0-9]+),([0-9]+)$|",$1,".log($2)/log(2)|e if ($flavour eq "macosx"); 540 } 541 $line = ""; 542 return $self; 543 } 544 545 # non-gas case or nasm/masm 546 SWITCH: for ($dir) { 547 /\.text/ && do { my $v=undef; 548 if ($nasm) { 549 $v="section .text code align=64\n"; 550 } else { 551 $v="$current_segment\tENDS\n" if ($current_segment); 552 $current_segment = ".text\$"; 553 $v.="$current_segment\tSEGMENT "; 554 $v.=$masm>=$masmref ? "ALIGN(256)" : "PAGE"; 555 $v.=" 'CODE'"; 556 } 557 $self->{value} = $v; 558 last; 559 }; 560 /\.data/ && do { my $v=undef; 561 if ($nasm) { 562 $v="section .data data align=8\n"; 563 } else { 564 $v="$current_segment\tENDS\n" if ($current_segment); 565 $current_segment = "_DATA"; 566 $v.="$current_segment\tSEGMENT"; 567 } 568 $self->{value} = $v; 569 last; 570 }; 571 /\.section/ && do { my $v=undef; 572 $line =~ s/([^,]*).*/$1/; 573 $line = ".CRT\$XCU" if ($line eq ".init"); 574 if ($nasm) { 575 $v="section $line"; 576 if ($line=~/\.([px])data/) { 577 $v.=" rdata align="; 578 $v.=$1 eq "p"? 4 : 8; 579 } elsif ($line=~/\.CRT\$/i) { 580 $v.=" rdata align=8"; 581 } 582 } else { 583 $v="$current_segment\tENDS\n" if ($current_segment); 584 $v.="$line\tSEGMENT"; 585 if ($line=~/\.([px])data/) { 586 $v.=" READONLY"; 587 $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref); 588 } elsif ($line=~/\.CRT\$/i) { 589 $v.=" READONLY "; 590 $v.=$masm>=$masmref ? "ALIGN(8)" : "DWORD"; 591 } 592 } 593 $current_segment = $line; 594 $self->{value} = $v; 595 last; 596 }; 597 /\.extern/ && do { $self->{value} = "EXTERN\t".$line; 598 $self->{value} .= ":NEAR" if ($masm); 599 last; 600 }; 601 /\.globl|.global/ 602 && do { $self->{value} = $masm?"PUBLIC":"global"; 603 $self->{value} .= "\t".$line; 604 last; 605 }; 606 /\.size/ && do { if (defined($current_function)) { 607 undef $self->{value}; 608 if ($current_function->{abi} eq "svr4") { 609 $self->{value}="${decor}SEH_end_$current_function->{name}:"; 610 $self->{value}.=":\n" if($masm); 611 } 612 $self->{value}.="$current_function->{name}\tENDP" if($masm && $current_function->{name}); 613 undef $current_function; 614 } 615 last; 616 }; 617 /\.align/ && do { $self->{value} = "ALIGN\t".$line; last; }; 618 /\.(value|long|rva|quad)/ 619 && do { my $sz = substr($1,0,1); 620 my @arr = split(/,\s*/,$line); 621 my $last = pop(@arr); 622 my $conv = sub { my $var=shift; 623 $var=~s/^(0b[0-1]+)/oct($1)/eig; 624 $var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm); 625 if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva")) 626 { $var=~s/([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; } 627 $var; 628 }; 629 630 $sz =~ tr/bvlrq/BWDDQ/; 631 $self->{value} = "\tD$sz\t"; 632 for (@arr) { $self->{value} .= &$conv($_).","; } 633 $self->{value} .= &$conv($last); 634 last; 635 }; 636 /\.byte/ && do { my @str=split(/,\s*/,$line); 637 map(s/(0b[0-1]+)/oct($1)/eig,@str); 638 map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm); 639 while ($#str>15) { 640 $self->{value}.="DB\t" 641 .join(",",@str[0..15])."\n"; 642 foreach (0..15) { shift @str; } 643 } 644 $self->{value}.="DB\t" 645 .join(",",@str) if (@str); 646 last; 647 }; 648 /\.comm/ && do { my @str=split(/,\s*/,$line); 649 my $v=undef; 650 if ($nasm) { 651 $v.="common $prefix@str[0] @str[1]"; 652 } else { 653 $v="$current_segment\tENDS\n" if ($current_segment); 654 $current_segment = "_DATA"; 655 $v.="$current_segment\tSEGMENT\n"; 656 $v.="COMM @str[0]:DWORD:".@str[1]/4; 657 } 658 $self->{value} = $v; 659 last; 660 }; 661 } 662 $line = ""; 663 } 664 665 $ret; 666 } 667 sub out { 668 my $self = shift; 669 $self->{value}; 670 } 671} 672 673sub rex { 674 local *opcode=shift; 675 my ($dst,$src,$rex)=@_; 676 677 $rex|=0x04 if($dst>=8); 678 $rex|=0x01 if($src>=8); 679 push @opcode,($rex|0x40) if ($rex); 680} 681 682# older gas and ml64 don't handle SSE>2 instructions 683my %regrm = ( "%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3, 684 "%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7 ); 685 686my $movq = sub { # elderly gas can't handle inter-register movq 687 my $arg = shift; 688 my @opcode=(0x66); 689 if ($arg =~ /%xmm([0-9]+),\s*%r(\w+)/) { 690 my ($src,$dst)=($1,$2); 691 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 692 rex(\@opcode,$src,$dst,0x8); 693 push @opcode,0x0f,0x7e; 694 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 695 @opcode; 696 } elsif ($arg =~ /%r(\w+),\s*%xmm([0-9]+)/) { 697 my ($src,$dst)=($2,$1); 698 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 699 rex(\@opcode,$src,$dst,0x8); 700 push @opcode,0x0f,0x6e; 701 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 702 @opcode; 703 } else { 704 (); 705 } 706}; 707 708my $pextrd = sub { 709 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*(%\w+)/) { 710 my @opcode=(0x66); 711 $imm=$1; 712 $src=$2; 713 $dst=$3; 714 if ($dst =~ /%r([0-9]+)d/) { $dst = $1; } 715 elsif ($dst =~ /%e/) { $dst = $regrm{$dst}; } 716 rex(\@opcode,$src,$dst); 717 push @opcode,0x0f,0x3a,0x16; 718 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 719 push @opcode,$imm; 720 @opcode; 721 } else { 722 (); 723 } 724}; 725 726my $pinsrd = sub { 727 if (shift =~ /\$([0-9]+),\s*(%\w+),\s*%xmm([0-9]+)/) { 728 my @opcode=(0x66); 729 $imm=$1; 730 $src=$2; 731 $dst=$3; 732 if ($src =~ /%r([0-9]+)/) { $src = $1; } 733 elsif ($src =~ /%e/) { $src = $regrm{$src}; } 734 rex(\@opcode,$dst,$src); 735 push @opcode,0x0f,0x3a,0x22; 736 push @opcode,0xc0|(($dst&7)<<3)|($src&7); # ModR/M 737 push @opcode,$imm; 738 @opcode; 739 } else { 740 (); 741 } 742}; 743 744my $pshufb = sub { 745 if (shift =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) { 746 my @opcode=(0x66); 747 rex(\@opcode,$2,$1); 748 push @opcode,0x0f,0x38,0x00; 749 push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M 750 @opcode; 751 } else { 752 (); 753 } 754}; 755 756my $palignr = sub { 757 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 758 my @opcode=(0x66); 759 rex(\@opcode,$3,$2); 760 push @opcode,0x0f,0x3a,0x0f; 761 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 762 push @opcode,$1; 763 @opcode; 764 } else { 765 (); 766 } 767}; 768 769my $pclmulqdq = sub { 770 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 771 my @opcode=(0x66); 772 rex(\@opcode,$3,$2); 773 push @opcode,0x0f,0x3a,0x44; 774 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 775 my $c=$1; 776 push @opcode,$c=~/^0/?oct($c):$c; 777 @opcode; 778 } else { 779 (); 780 } 781}; 782 783my $rdrand = sub { 784 if (shift =~ /%[er](\w+)/) { 785 my @opcode=(); 786 my $dst=$1; 787 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 788 rex(\@opcode,0,$1,8); 789 push @opcode,0x0f,0xc7,0xf0|($dst&7); 790 @opcode; 791 } else { 792 (); 793 } 794}; 795 796my $rdseed = sub { 797 if (shift =~ /%[er](\w+)/) { 798 my @opcode=(); 799 my $dst=$1; 800 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 801 rex(\@opcode,0,$1,8); 802 push @opcode,0x0f,0xc7,0xf8|($dst&7); 803 @opcode; 804 } else { 805 (); 806 } 807}; 808 809sub rxb { 810 local *opcode=shift; 811 my ($dst,$src1,$src2,$rxb)=@_; 812 813 $rxb|=0x7<<5; 814 $rxb&=~(0x04<<5) if($dst>=8); 815 $rxb&=~(0x01<<5) if($src1>=8); 816 $rxb&=~(0x02<<5) if($src2>=8); 817 push @opcode,$rxb; 818} 819 820my $vprotd = sub { 821 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 822 my @opcode=(0x8f); 823 rxb(\@opcode,$3,$2,-1,0x08); 824 push @opcode,0x78,0xc2; 825 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 826 my $c=$1; 827 push @opcode,$c=~/^0/?oct($c):$c; 828 @opcode; 829 } else { 830 (); 831 } 832}; 833 834my $vprotq = sub { 835 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 836 my @opcode=(0x8f); 837 rxb(\@opcode,$3,$2,-1,0x08); 838 push @opcode,0x78,0xc3; 839 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 840 my $c=$1; 841 push @opcode,$c=~/^0/?oct($c):$c; 842 @opcode; 843 } else { 844 (); 845 } 846}; 847 848if ($nasm) { 849 print <<___; 850default rel 851%define XMMWORD 852%define YMMWORD 853%define ZMMWORD 854___ 855} elsif ($masm) { 856 print <<___; 857OPTION DOTNAME 858___ 859} 860while($line=<>) { 861 862 chomp($line); 863 864 $line =~ s|[#!].*$||; # get rid of asm-style comments... 865 $line =~ s|/\*.*\*/||; # ... and C-style comments... 866 $line =~ s|^\s+||; # ... and skip white spaces in beginning 867 $line =~ s|\s+$||; # ... and at the end 868 869 undef $label; 870 undef $opcode; 871 undef @args; 872 873 if ($label=label->re(\$line)) { print $label->out(); } 874 875 if (directive->re(\$line)) { 876 printf "%s",directive->out(); 877 } elsif ($opcode=opcode->re(\$line)) { 878 my $asm = eval("\$".$opcode->mnemonic()); 879 undef @bytes; 880 881 if ((ref($asm) eq 'CODE') && scalar(@bytes=&$asm($line))) { 882 print $gas?".byte\t":"DB\t",join(',',@bytes),"\n"; 883 next; 884 } 885 886 ARGUMENT: while (1) { 887 my $arg; 888 889 if ($arg=register->re(\$line)) { opcode->size($arg->size()); } 890 elsif ($arg=const->re(\$line)) { } 891 elsif ($arg=ea->re(\$line)) { } 892 elsif ($arg=expr->re(\$line)) { } 893 else { last ARGUMENT; } 894 895 push @args,$arg; 896 897 last ARGUMENT if ($line !~ /^,/); 898 899 $line =~ s/^,\s*//; 900 } # ARGUMENT: 901 902 if ($#args>=0) { 903 my $insn; 904 my $sz=opcode->size(); 905 906 if ($gas) { 907 $insn = $opcode->out($#args>=1?$args[$#args]->size():$sz); 908 @args = map($_->out($sz),@args); 909 printf "\t%s\t%s",$insn,join(",",@args); 910 } else { 911 $insn = $opcode->out(); 912 foreach (@args) { 913 my $arg = $_->out(); 914 # $insn.=$sz compensates for movq, pinsrw, ... 915 if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; } 916 if ($arg =~ /^ymm[0-9]+$/) { $insn.=$sz; $sz="y" if(!$sz); last; } 917 if ($arg =~ /^zmm[0-9]+$/) { $insn.=$sz; $sz="z" if(!$sz); last; } 918 if ($arg =~ /^mm[0-9]+$/) { $insn.=$sz; $sz="q" if(!$sz); last; } 919 } 920 @args = reverse(@args); 921 undef $sz if ($nasm && $opcode->mnemonic() eq "lea"); 922 printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args)); 923 } 924 } else { 925 printf "\t%s",$opcode->out(); 926 } 927 } 928 929 print $line,"\n"; 930} 931 932print "\n$current_segment\tENDS\n" if ($current_segment && $masm); 933print "END\n" if ($masm); 934 935close STDOUT; 936 937################################################# 938# Cross-reference x86_64 ABI "card" 939# 940# Unix Win64 941# %rax * * 942# %rbx - - 943# %rcx #4 #1 944# %rdx #3 #2 945# %rsi #2 - 946# %rdi #1 - 947# %rbp - - 948# %rsp - - 949# %r8 #5 #3 950# %r9 #6 #4 951# %r10 * * 952# %r11 * * 953# %r12 - - 954# %r13 - - 955# %r14 - - 956# %r15 - - 957# 958# (*) volatile register 959# (-) preserved by callee 960# (#) Nth argument, volatile 961# 962# In Unix terms top of stack is argument transfer area for arguments 963# which could not be accomodated in registers. Or in other words 7th 964# [integer] argument resides at 8(%rsp) upon function entry point. 965# 128 bytes above %rsp constitute a "red zone" which is not touched 966# by signal handlers and can be used as temporal storage without 967# allocating a frame. 968# 969# In Win64 terms N*8 bytes on top of stack is argument transfer area, 970# which belongs to/can be overwritten by callee. N is the number of 971# arguments passed to callee, *but* not less than 4! This means that 972# upon function entry point 5th argument resides at 40(%rsp), as well 973# as that 32 bytes from 8(%rsp) can always be used as temporal 974# storage [without allocating a frame]. One can actually argue that 975# one can assume a "red zone" above stack pointer under Win64 as well. 976# Point is that at apparently no occasion Windows kernel would alter 977# the area above user stack pointer in true asynchronous manner... 978# 979# All the above means that if assembler programmer adheres to Unix 980# register and stack layout, but disregards the "red zone" existense, 981# it's possible to use following prologue and epilogue to "gear" from 982# Unix to Win64 ABI in leaf functions with not more than 6 arguments. 983# 984# omnipotent_function: 985# ifdef WIN64 986# movq %rdi,8(%rsp) 987# movq %rsi,16(%rsp) 988# movq %rcx,%rdi ; if 1st argument is actually present 989# movq %rdx,%rsi ; if 2nd argument is actually ... 990# movq %r8,%rdx ; if 3rd argument is ... 991# movq %r9,%rcx ; if 4th argument ... 992# movq 40(%rsp),%r8 ; if 5th ... 993# movq 48(%rsp),%r9 ; if 6th ... 994# endif 995# ... 996# ifdef WIN64 997# movq 8(%rsp),%rdi 998# movq 16(%rsp),%rsi 999# endif 1000# ret 1001# 1002################################################# 1003# Win64 SEH, Structured Exception Handling. 1004# 1005# Unlike on Unix systems(*) lack of Win64 stack unwinding information 1006# has undesired side-effect at run-time: if an exception is raised in 1007# assembler subroutine such as those in question (basically we're 1008# referring to segmentation violations caused by malformed input 1009# parameters), the application is briskly terminated without invoking 1010# any exception handlers, most notably without generating memory dump 1011# or any user notification whatsoever. This poses a problem. It's 1012# possible to address it by registering custom language-specific 1013# handler that would restore processor context to the state at 1014# subroutine entry point and return "exception is not handled, keep 1015# unwinding" code. Writing such handler can be a challenge... But it's 1016# doable, though requires certain coding convention. Consider following 1017# snippet: 1018# 1019# .type function,@function 1020# function: 1021# movq %rsp,%rax # copy rsp to volatile register 1022# pushq %r15 # save non-volatile registers 1023# pushq %rbx 1024# pushq %rbp 1025# movq %rsp,%r11 1026# subq %rdi,%r11 # prepare [variable] stack frame 1027# andq $-64,%r11 1028# movq %rax,0(%r11) # check for exceptions 1029# movq %r11,%rsp # allocate [variable] stack frame 1030# movq %rax,0(%rsp) # save original rsp value 1031# magic_point: 1032# ... 1033# movq 0(%rsp),%rcx # pull original rsp value 1034# movq -24(%rcx),%rbp # restore non-volatile registers 1035# movq -16(%rcx),%rbx 1036# movq -8(%rcx),%r15 1037# movq %rcx,%rsp # restore original rsp 1038# ret 1039# .size function,.-function 1040# 1041# The key is that up to magic_point copy of original rsp value remains 1042# in chosen volatile register and no non-volatile register, except for 1043# rsp, is modified. While past magic_point rsp remains constant till 1044# the very end of the function. In this case custom language-specific 1045# exception handler would look like this: 1046# 1047# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, 1048# CONTEXT *context,DISPATCHER_CONTEXT *disp) 1049# { ULONG64 *rsp = (ULONG64 *)context->Rax; 1050# if (context->Rip >= magic_point) 1051# { rsp = ((ULONG64 **)context->Rsp)[0]; 1052# context->Rbp = rsp[-3]; 1053# context->Rbx = rsp[-2]; 1054# context->R15 = rsp[-1]; 1055# } 1056# context->Rsp = (ULONG64)rsp; 1057# context->Rdi = rsp[1]; 1058# context->Rsi = rsp[2]; 1059# 1060# memcpy (disp->ContextRecord,context,sizeof(CONTEXT)); 1061# RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase, 1062# dips->ControlPc,disp->FunctionEntry,disp->ContextRecord, 1063# &disp->HandlerData,&disp->EstablisherFrame,NULL); 1064# return ExceptionContinueSearch; 1065# } 1066# 1067# It's appropriate to implement this handler in assembler, directly in 1068# function's module. In order to do that one has to know members' 1069# offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant 1070# values. Here they are: 1071# 1072# CONTEXT.Rax 120 1073# CONTEXT.Rcx 128 1074# CONTEXT.Rdx 136 1075# CONTEXT.Rbx 144 1076# CONTEXT.Rsp 152 1077# CONTEXT.Rbp 160 1078# CONTEXT.Rsi 168 1079# CONTEXT.Rdi 176 1080# CONTEXT.R8 184 1081# CONTEXT.R9 192 1082# CONTEXT.R10 200 1083# CONTEXT.R11 208 1084# CONTEXT.R12 216 1085# CONTEXT.R13 224 1086# CONTEXT.R14 232 1087# CONTEXT.R15 240 1088# CONTEXT.Rip 248 1089# CONTEXT.Xmm6 512 1090# sizeof(CONTEXT) 1232 1091# DISPATCHER_CONTEXT.ControlPc 0 1092# DISPATCHER_CONTEXT.ImageBase 8 1093# DISPATCHER_CONTEXT.FunctionEntry 16 1094# DISPATCHER_CONTEXT.EstablisherFrame 24 1095# DISPATCHER_CONTEXT.TargetIp 32 1096# DISPATCHER_CONTEXT.ContextRecord 40 1097# DISPATCHER_CONTEXT.LanguageHandler 48 1098# DISPATCHER_CONTEXT.HandlerData 56 1099# UNW_FLAG_NHANDLER 0 1100# ExceptionContinueSearch 1 1101# 1102# In order to tie the handler to the function one has to compose 1103# couple of structures: one for .xdata segment and one for .pdata. 1104# 1105# UNWIND_INFO structure for .xdata segment would be 1106# 1107# function_unwind_info: 1108# .byte 9,0,0,0 1109# .rva handler 1110# 1111# This structure designates exception handler for a function with 1112# zero-length prologue, no stack frame or frame register. 1113# 1114# To facilitate composing of .pdata structures, auto-generated "gear" 1115# prologue copies rsp value to rax and denotes next instruction with 1116# .LSEH_begin_{function_name} label. This essentially defines the SEH 1117# styling rule mentioned in the beginning. Position of this label is 1118# chosen in such manner that possible exceptions raised in the "gear" 1119# prologue would be accounted to caller and unwound from latter's frame. 1120# End of function is marked with respective .LSEH_end_{function_name} 1121# label. To summarize, .pdata segment would contain 1122# 1123# .rva .LSEH_begin_function 1124# .rva .LSEH_end_function 1125# .rva function_unwind_info 1126# 1127# Reference to functon_unwind_info from .xdata segment is the anchor. 1128# In case you wonder why references are 32-bit .rvas and not 64-bit 1129# .quads. References put into these two segments are required to be 1130# *relative* to the base address of the current binary module, a.k.a. 1131# image base. No Win64 module, be it .exe or .dll, can be larger than 1132# 2GB and thus such relative references can be and are accommodated in 1133# 32 bits. 1134# 1135# Having reviewed the example function code, one can argue that "movq 1136# %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix 1137# rax would contain an undefined value. If this "offends" you, use 1138# another register and refrain from modifying rax till magic_point is 1139# reached, i.e. as if it was a non-volatile register. If more registers 1140# are required prior [variable] frame setup is completed, note that 1141# nobody says that you can have only one "magic point." You can 1142# "liberate" non-volatile registers by denoting last stack off-load 1143# instruction and reflecting it in finer grade unwind logic in handler. 1144# After all, isn't it why it's called *language-specific* handler... 1145# 1146# Attentive reader can notice that exceptions would be mishandled in 1147# auto-generated "gear" epilogue. Well, exception effectively can't 1148# occur there, because if memory area used by it was subject to 1149# segmentation violation, then it would be raised upon call to the 1150# function (and as already mentioned be accounted to caller, which is 1151# not a problem). If you're still not comfortable, then define tail 1152# "magic point" just prior ret instruction and have handler treat it... 1153# 1154# (*) Note that we're talking about run-time, not debug-time. Lack of 1155# unwind information makes debugging hard on both Windows and 1156# Unix. "Unlike" referes to the fact that on Unix signal handler 1157# will always be invoked, core dumped and appropriate exit code 1158# returned to parent (for user notification). 1159