x86_64-xlate.pl revision 296341
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/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg; 254 $self->{label} =~ s/([0-9]+)/$1<<32>>32/eg; 255 256 if ($gas) { 257 $self->{label} =~ s/^___imp_/__imp__/ if ($flavour eq "mingw64"); 258 259 if (defined($self->{index})) { 260 sprintf "%s%s(%s,%%%s,%d)",$self->{asterisk}, 261 $self->{label}, 262 $self->{base}?"%$self->{base}":"", 263 $self->{index},$self->{scale}; 264 } else { 265 sprintf "%s%s(%%%s)", $self->{asterisk},$self->{label},$self->{base}; 266 } 267 } else { 268 %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR", l=>"DWORD$PTR", 269 q=>"QWORD$PTR",o=>"OWORD$PTR",x=>"XMMWORD$PTR" ); 270 271 $self->{label} =~ s/\./\$/g; 272 $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig; 273 $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/); 274 $sz="q" if ($self->{asterisk} || opcode->mnemonic() eq "movq"); 275 $sz="l" if (opcode->mnemonic() eq "movd"); 276 277 if (defined($self->{index})) { 278 sprintf "%s[%s%s*%d%s]",$szmap{$sz}, 279 $self->{label}?"$self->{label}+":"", 280 $self->{index},$self->{scale}, 281 $self->{base}?"+$self->{base}":""; 282 } elsif ($self->{base} eq "rip") { 283 sprintf "%s[%s]",$szmap{$sz},$self->{label}; 284 } else { 285 sprintf "%s[%s%s]",$szmap{$sz}, 286 $self->{label}?"$self->{label}+":"", 287 $self->{base}; 288 } 289 } 290 } 291} 292{ package register; # pick up registers, which start with %. 293 sub re { 294 my $class = shift; # muliple instances... 295 my $self = {}; 296 local *line = shift; 297 undef $ret; 298 299 # optional * ---vvv--- appears in indirect jmp/call 300 if ($line =~ /^(\*?)%(\w+)/) { 301 bless $self,$class; 302 $self->{asterisk} = $1; 303 $self->{value} = $2; 304 $ret = $self; 305 $line = substr($line,@+[0]); $line =~ s/^\s+//; 306 } 307 $ret; 308 } 309 sub size { 310 my $self = shift; 311 undef $ret; 312 313 if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; } 314 elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; } 315 elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; } 316 elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; } 317 elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; } 318 elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; } 319 elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; } 320 elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; } 321 322 $ret; 323 } 324 sub out { 325 my $self = shift; 326 if ($gas) { sprintf "%s%%%s",$self->{asterisk},$self->{value}; } 327 else { $self->{value}; } 328 } 329} 330{ package label; # pick up labels, which end with : 331 sub re { 332 my $self = shift; # single instance is enough... 333 local *line = shift; 334 undef $ret; 335 336 if ($line =~ /(^[\.\w]+)\:/) { 337 $self->{value} = $1; 338 $ret = $self; 339 $line = substr($line,@+[0]); $line =~ s/^\s+//; 340 341 $self->{value} =~ s/^\.L/$decor/; 342 } 343 $ret; 344 } 345 sub out { 346 my $self = shift; 347 348 if ($gas) { 349 my $func = ($globals{$self->{value}} or $self->{value}) . ":"; 350 if ($win64 && 351 $current_function->{name} eq $self->{value} && 352 $current_function->{abi} eq "svr4") { 353 $func .= "\n"; 354 $func .= " movq %rdi,8(%rsp)\n"; 355 $func .= " movq %rsi,16(%rsp)\n"; 356 $func .= " movq %rsp,%rax\n"; 357 $func .= "${decor}SEH_begin_$current_function->{name}:\n"; 358 my $narg = $current_function->{narg}; 359 $narg=6 if (!defined($narg)); 360 $func .= " movq %rcx,%rdi\n" if ($narg>0); 361 $func .= " movq %rdx,%rsi\n" if ($narg>1); 362 $func .= " movq %r8,%rdx\n" if ($narg>2); 363 $func .= " movq %r9,%rcx\n" if ($narg>3); 364 $func .= " movq 40(%rsp),%r8\n" if ($narg>4); 365 $func .= " movq 48(%rsp),%r9\n" if ($narg>5); 366 } 367 $func; 368 } elsif ($self->{value} ne "$current_function->{name}") { 369 $self->{value} .= ":" if ($masm && $ret!~m/^\$/); 370 $self->{value} . ":"; 371 } elsif ($win64 && $current_function->{abi} eq "svr4") { 372 my $func = "$current_function->{name}" . 373 ($nasm ? ":" : "\tPROC $current_function->{scope}") . 374 "\n"; 375 $func .= " mov QWORD${PTR}[8+rsp],rdi\t;WIN64 prologue\n"; 376 $func .= " mov QWORD${PTR}[16+rsp],rsi\n"; 377 $func .= " mov rax,rsp\n"; 378 $func .= "${decor}SEH_begin_$current_function->{name}:"; 379 $func .= ":" if ($masm); 380 $func .= "\n"; 381 my $narg = $current_function->{narg}; 382 $narg=6 if (!defined($narg)); 383 $func .= " mov rdi,rcx\n" if ($narg>0); 384 $func .= " mov rsi,rdx\n" if ($narg>1); 385 $func .= " mov rdx,r8\n" if ($narg>2); 386 $func .= " mov rcx,r9\n" if ($narg>3); 387 $func .= " mov r8,QWORD${PTR}[40+rsp]\n" if ($narg>4); 388 $func .= " mov r9,QWORD${PTR}[48+rsp]\n" if ($narg>5); 389 $func .= "\n"; 390 } else { 391 "$current_function->{name}". 392 ($nasm ? ":" : "\tPROC $current_function->{scope}"); 393 } 394 } 395} 396{ package expr; # pick up expressioins 397 sub re { 398 my $self = shift; # single instance is enough... 399 local *line = shift; 400 undef $ret; 401 402 if ($line =~ /(^[^,]+)/) { 403 $self->{value} = $1; 404 $ret = $self; 405 $line = substr($line,@+[0]); $line =~ s/^\s+//; 406 407 $self->{value} =~ s/\@PLT// if (!$elf); 408 $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 409 $self->{value} =~ s/\.L/$decor/g; 410 } 411 $ret; 412 } 413 sub out { 414 my $self = shift; 415 if ($nasm && opcode->mnemonic()=~m/^j/) { 416 "NEAR ".$self->{value}; 417 } else { 418 $self->{value}; 419 } 420 } 421} 422{ package directive; # pick up directives, which start with . 423 sub re { 424 my $self = shift; # single instance is enough... 425 local *line = shift; 426 undef $ret; 427 my $dir; 428 my %opcode = # lea 2f-1f(%rip),%dst; 1: nop; 2: 429 ( "%rax"=>0x01058d48, "%rcx"=>0x010d8d48, 430 "%rdx"=>0x01158d48, "%rbx"=>0x011d8d48, 431 "%rsp"=>0x01258d48, "%rbp"=>0x012d8d48, 432 "%rsi"=>0x01358d48, "%rdi"=>0x013d8d48, 433 "%r8" =>0x01058d4c, "%r9" =>0x010d8d4c, 434 "%r10"=>0x01158d4c, "%r11"=>0x011d8d4c, 435 "%r12"=>0x01258d4c, "%r13"=>0x012d8d4c, 436 "%r14"=>0x01358d4c, "%r15"=>0x013d8d4c ); 437 438 if ($line =~ /^\s*(\.\w+)/) { 439 $dir = $1; 440 $ret = $self; 441 undef $self->{value}; 442 $line = substr($line,@+[0]); $line =~ s/^\s+//; 443 444 SWITCH: for ($dir) { 445 /\.picmeup/ && do { if ($line =~ /(%r[\w]+)/i) { 446 $dir="\t.long"; 447 $line=sprintf "0x%x,0x90000000",$opcode{$1}; 448 } 449 last; 450 }; 451 /\.global|\.globl|\.extern/ 452 && do { $globals{$line} = $prefix . $line; 453 $line = $globals{$line} if ($prefix); 454 last; 455 }; 456 /\.type/ && do { ($sym,$type,$narg) = split(',',$line); 457 if ($type eq "\@function") { 458 undef $current_function; 459 $current_function->{name} = $sym; 460 $current_function->{abi} = "svr4"; 461 $current_function->{narg} = $narg; 462 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 463 } elsif ($type eq "\@abi-omnipotent") { 464 undef $current_function; 465 $current_function->{name} = $sym; 466 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 467 } 468 $line =~ s/\@abi\-omnipotent/\@function/; 469 $line =~ s/\@function.*/\@function/; 470 last; 471 }; 472 /\.asciz/ && do { if ($line =~ /^"(.*)"$/) { 473 $dir = ".byte"; 474 $line = join(",",unpack("C*",$1),0); 475 } 476 last; 477 }; 478 /\.rva|\.long|\.quad/ 479 && do { $line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 480 $line =~ s/\.L/$decor/g; 481 last; 482 }; 483 } 484 485 if ($gas) { 486 $self->{value} = $dir . "\t" . $line; 487 488 if ($dir =~ /\.extern/) { 489 $self->{value} = ""; # swallow extern 490 } elsif (!$elf && $dir =~ /\.type/) { 491 $self->{value} = ""; 492 $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" . 493 (defined($globals{$1})?".scl 2;":".scl 3;") . 494 "\t.type 32;\t.endef" 495 if ($win64 && $line =~ /([^,]+),\@function/); 496 } elsif (!$elf && $dir =~ /\.size/) { 497 $self->{value} = ""; 498 if (defined($current_function)) { 499 $self->{value} .= "${decor}SEH_end_$current_function->{name}:" 500 if ($win64 && $current_function->{abi} eq "svr4"); 501 undef $current_function; 502 } 503 } elsif (!$elf && $dir =~ /\.align/) { 504 $self->{value} = ".p2align\t" . (log($line)/log(2)); 505 } elsif ($dir eq ".section") { 506 $current_segment=$line; 507 if (!$elf && $current_segment eq ".init") { 508 if ($flavour eq "macosx") { $self->{value} = ".mod_init_func"; } 509 elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.ctors"; } 510 } 511 } elsif ($dir =~ /\.(text|data)/) { 512 $current_segment=".$1"; 513 } elsif ($dir =~ /\.hidden/) { 514 if ($flavour eq "macosx") { $self->{value} = ".private_extern\t$prefix$line"; } 515 elsif ($flavour eq "mingw64") { $self->{value} = ""; } 516 } elsif ($dir =~ /\.comm/) { 517 $self->{value} = "$dir\t$prefix$line"; 518 $self->{value} =~ s|,([0-9]+),([0-9]+)$|",$1,".log($2)/log(2)|e if ($flavour eq "macosx"); 519 } 520 $line = ""; 521 return $self; 522 } 523 524 # non-gas case or nasm/masm 525 SWITCH: for ($dir) { 526 /\.text/ && do { my $v=undef; 527 if ($nasm) { 528 $v="section .text code align=64\n"; 529 } else { 530 $v="$current_segment\tENDS\n" if ($current_segment); 531 $current_segment = ".text\$"; 532 $v.="$current_segment\tSEGMENT "; 533 $v.=$masm>=$masmref ? "ALIGN(64)" : "PAGE"; 534 $v.=" 'CODE'"; 535 } 536 $self->{value} = $v; 537 last; 538 }; 539 /\.data/ && do { my $v=undef; 540 if ($nasm) { 541 $v="section .data data align=8\n"; 542 } else { 543 $v="$current_segment\tENDS\n" if ($current_segment); 544 $current_segment = "_DATA"; 545 $v.="$current_segment\tSEGMENT"; 546 } 547 $self->{value} = $v; 548 last; 549 }; 550 /\.section/ && do { my $v=undef; 551 $line =~ s/([^,]*).*/$1/; 552 $line = ".CRT\$XCU" if ($line eq ".init"); 553 if ($nasm) { 554 $v="section $line"; 555 if ($line=~/\.([px])data/) { 556 $v.=" rdata align="; 557 $v.=$1 eq "p"? 4 : 8; 558 } elsif ($line=~/\.CRT\$/i) { 559 $v.=" rdata align=8"; 560 } 561 } else { 562 $v="$current_segment\tENDS\n" if ($current_segment); 563 $v.="$line\tSEGMENT"; 564 if ($line=~/\.([px])data/) { 565 $v.=" READONLY"; 566 $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref); 567 } elsif ($line=~/\.CRT\$/i) { 568 $v.=" READONLY "; 569 $v.=$masm>=$masmref ? "ALIGN(8)" : "DWORD"; 570 } 571 } 572 $current_segment = $line; 573 $self->{value} = $v; 574 last; 575 }; 576 /\.extern/ && do { $self->{value} = "EXTERN\t".$line; 577 $self->{value} .= ":NEAR" if ($masm); 578 last; 579 }; 580 /\.globl|.global/ 581 && do { $self->{value} = $masm?"PUBLIC":"global"; 582 $self->{value} .= "\t".$line; 583 last; 584 }; 585 /\.size/ && do { if (defined($current_function)) { 586 undef $self->{value}; 587 if ($current_function->{abi} eq "svr4") { 588 $self->{value}="${decor}SEH_end_$current_function->{name}:"; 589 $self->{value}.=":\n" if($masm); 590 } 591 $self->{value}.="$current_function->{name}\tENDP" if($masm && $current_function->{name}); 592 undef $current_function; 593 } 594 last; 595 }; 596 /\.align/ && do { $self->{value} = "ALIGN\t".$line; last; }; 597 /\.(value|long|rva|quad)/ 598 && do { my $sz = substr($1,0,1); 599 my @arr = split(/,\s*/,$line); 600 my $last = pop(@arr); 601 my $conv = sub { my $var=shift; 602 $var=~s/^(0b[0-1]+)/oct($1)/eig; 603 $var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm); 604 if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva")) 605 { $var=~s/([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; } 606 $var; 607 }; 608 609 $sz =~ tr/bvlrq/BWDDQ/; 610 $self->{value} = "\tD$sz\t"; 611 for (@arr) { $self->{value} .= &$conv($_).","; } 612 $self->{value} .= &$conv($last); 613 last; 614 }; 615 /\.byte/ && do { my @str=split(/,\s*/,$line); 616 map(s/(0b[0-1]+)/oct($1)/eig,@str); 617 map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm); 618 while ($#str>15) { 619 $self->{value}.="DB\t" 620 .join(",",@str[0..15])."\n"; 621 foreach (0..15) { shift @str; } 622 } 623 $self->{value}.="DB\t" 624 .join(",",@str) if (@str); 625 last; 626 }; 627 /\.comm/ && do { my @str=split(/,\s*/,$line); 628 my $v=undef; 629 if ($nasm) { 630 $v.="common $prefix@str[0] @str[1]"; 631 } else { 632 $v="$current_segment\tENDS\n" if ($current_segment); 633 $current_segment = "_DATA"; 634 $v.="$current_segment\tSEGMENT\n"; 635 $v.="COMM @str[0]:DWORD:".@str[1]/4; 636 } 637 $self->{value} = $v; 638 last; 639 }; 640 } 641 $line = ""; 642 } 643 644 $ret; 645 } 646 sub out { 647 my $self = shift; 648 $self->{value}; 649 } 650} 651 652sub rex { 653 local *opcode=shift; 654 my ($dst,$src,$rex)=@_; 655 656 $rex|=0x04 if($dst>=8); 657 $rex|=0x01 if($src>=8); 658 push @opcode,($rex|0x40) if ($rex); 659} 660 661# older gas and ml64 don't handle SSE>2 instructions 662my %regrm = ( "%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3, 663 "%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7 ); 664 665my $movq = sub { # elderly gas can't handle inter-register movq 666 my $arg = shift; 667 my @opcode=(0x66); 668 if ($arg =~ /%xmm([0-9]+),\s*%r(\w+)/) { 669 my ($src,$dst)=($1,$2); 670 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 671 rex(\@opcode,$src,$dst,0x8); 672 push @opcode,0x0f,0x7e; 673 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 674 @opcode; 675 } elsif ($arg =~ /%r(\w+),\s*%xmm([0-9]+)/) { 676 my ($src,$dst)=($2,$1); 677 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 678 rex(\@opcode,$src,$dst,0x8); 679 push @opcode,0x0f,0x6e; 680 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 681 @opcode; 682 } else { 683 (); 684 } 685}; 686 687my $pextrd = sub { 688 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*(%\w+)/) { 689 my @opcode=(0x66); 690 $imm=$1; 691 $src=$2; 692 $dst=$3; 693 if ($dst =~ /%r([0-9]+)d/) { $dst = $1; } 694 elsif ($dst =~ /%e/) { $dst = $regrm{$dst}; } 695 rex(\@opcode,$src,$dst); 696 push @opcode,0x0f,0x3a,0x16; 697 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 698 push @opcode,$imm; 699 @opcode; 700 } else { 701 (); 702 } 703}; 704 705my $pinsrd = sub { 706 if (shift =~ /\$([0-9]+),\s*(%\w+),\s*%xmm([0-9]+)/) { 707 my @opcode=(0x66); 708 $imm=$1; 709 $src=$2; 710 $dst=$3; 711 if ($src =~ /%r([0-9]+)/) { $src = $1; } 712 elsif ($src =~ /%e/) { $src = $regrm{$src}; } 713 rex(\@opcode,$dst,$src); 714 push @opcode,0x0f,0x3a,0x22; 715 push @opcode,0xc0|(($dst&7)<<3)|($src&7); # ModR/M 716 push @opcode,$imm; 717 @opcode; 718 } else { 719 (); 720 } 721}; 722 723my $pshufb = sub { 724 if (shift =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) { 725 my @opcode=(0x66); 726 rex(\@opcode,$2,$1); 727 push @opcode,0x0f,0x38,0x00; 728 push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M 729 @opcode; 730 } else { 731 (); 732 } 733}; 734 735my $palignr = sub { 736 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 737 my @opcode=(0x66); 738 rex(\@opcode,$3,$2); 739 push @opcode,0x0f,0x3a,0x0f; 740 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 741 push @opcode,$1; 742 @opcode; 743 } else { 744 (); 745 } 746}; 747 748my $pclmulqdq = sub { 749 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 750 my @opcode=(0x66); 751 rex(\@opcode,$3,$2); 752 push @opcode,0x0f,0x3a,0x44; 753 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 754 my $c=$1; 755 push @opcode,$c=~/^0/?oct($c):$c; 756 @opcode; 757 } else { 758 (); 759 } 760}; 761 762my $rdrand = sub { 763 if (shift =~ /%[er](\w+)/) { 764 my @opcode=(); 765 my $dst=$1; 766 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 767 rex(\@opcode,0,$1,8); 768 push @opcode,0x0f,0xc7,0xf0|($dst&7); 769 @opcode; 770 } else { 771 (); 772 } 773}; 774 775if ($nasm) { 776 print <<___; 777default rel 778%define XMMWORD 779___ 780} elsif ($masm) { 781 print <<___; 782OPTION DOTNAME 783___ 784} 785while($line=<>) { 786 787 chomp($line); 788 789 $line =~ s|[#!].*$||; # get rid of asm-style comments... 790 $line =~ s|/\*.*\*/||; # ... and C-style comments... 791 $line =~ s|^\s+||; # ... and skip white spaces in beginning 792 793 undef $label; 794 undef $opcode; 795 undef @args; 796 797 if ($label=label->re(\$line)) { print $label->out(); } 798 799 if (directive->re(\$line)) { 800 printf "%s",directive->out(); 801 } elsif ($opcode=opcode->re(\$line)) { 802 my $asm = eval("\$".$opcode->mnemonic()); 803 undef @bytes; 804 805 if ((ref($asm) eq 'CODE') && scalar(@bytes=&$asm($line))) { 806 print $gas?".byte\t":"DB\t",join(',',@bytes),"\n"; 807 next; 808 } 809 810 ARGUMENT: while (1) { 811 my $arg; 812 813 if ($arg=register->re(\$line)) { opcode->size($arg->size()); } 814 elsif ($arg=const->re(\$line)) { } 815 elsif ($arg=ea->re(\$line)) { } 816 elsif ($arg=expr->re(\$line)) { } 817 else { last ARGUMENT; } 818 819 push @args,$arg; 820 821 last ARGUMENT if ($line !~ /^,/); 822 823 $line =~ s/^,\s*//; 824 } # ARGUMENT: 825 826 if ($#args>=0) { 827 my $insn; 828 my $sz=opcode->size(); 829 830 if ($gas) { 831 $insn = $opcode->out($#args>=1?$args[$#args]->size():$sz); 832 @args = map($_->out($sz),@args); 833 printf "\t%s\t%s",$insn,join(",",@args); 834 } else { 835 $insn = $opcode->out(); 836 foreach (@args) { 837 my $arg = $_->out(); 838 # $insn.=$sz compensates for movq, pinsrw, ... 839 if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; } 840 if ($arg =~ /^mm[0-9]+$/) { $insn.=$sz; $sz="q" if(!$sz); last; } 841 } 842 @args = reverse(@args); 843 undef $sz if ($nasm && $opcode->mnemonic() eq "lea"); 844 printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args)); 845 } 846 } else { 847 printf "\t%s",$opcode->out(); 848 } 849 } 850 851 print $line,"\n"; 852} 853 854print "\n$current_segment\tENDS\n" if ($current_segment && $masm); 855print "END\n" if ($masm); 856 857close STDOUT; 858 859 ################################################# 860# Cross-reference x86_64 ABI "card" 861# 862# Unix Win64 863# %rax * * 864# %rbx - - 865# %rcx #4 #1 866# %rdx #3 #2 867# %rsi #2 - 868# %rdi #1 - 869# %rbp - - 870# %rsp - - 871# %r8 #5 #3 872# %r9 #6 #4 873# %r10 * * 874# %r11 * * 875# %r12 - - 876# %r13 - - 877# %r14 - - 878# %r15 - - 879# 880# (*) volatile register 881# (-) preserved by callee 882# (#) Nth argument, volatile 883# 884# In Unix terms top of stack is argument transfer area for arguments 885# which could not be accomodated in registers. Or in other words 7th 886# [integer] argument resides at 8(%rsp) upon function entry point. 887# 128 bytes above %rsp constitute a "red zone" which is not touched 888# by signal handlers and can be used as temporal storage without 889# allocating a frame. 890# 891# In Win64 terms N*8 bytes on top of stack is argument transfer area, 892# which belongs to/can be overwritten by callee. N is the number of 893# arguments passed to callee, *but* not less than 4! This means that 894# upon function entry point 5th argument resides at 40(%rsp), as well 895# as that 32 bytes from 8(%rsp) can always be used as temporal 896# storage [without allocating a frame]. One can actually argue that 897# one can assume a "red zone" above stack pointer under Win64 as well. 898# Point is that at apparently no occasion Windows kernel would alter 899# the area above user stack pointer in true asynchronous manner... 900# 901# All the above means that if assembler programmer adheres to Unix 902# register and stack layout, but disregards the "red zone" existense, 903# it's possible to use following prologue and epilogue to "gear" from 904# Unix to Win64 ABI in leaf functions with not more than 6 arguments. 905# 906# omnipotent_function: 907# ifdef WIN64 908# movq %rdi,8(%rsp) 909# movq %rsi,16(%rsp) 910# movq %rcx,%rdi ; if 1st argument is actually present 911# movq %rdx,%rsi ; if 2nd argument is actually ... 912# movq %r8,%rdx ; if 3rd argument is ... 913# movq %r9,%rcx ; if 4th argument ... 914# movq 40(%rsp),%r8 ; if 5th ... 915# movq 48(%rsp),%r9 ; if 6th ... 916# endif 917# ... 918# ifdef WIN64 919# movq 8(%rsp),%rdi 920# movq 16(%rsp),%rsi 921# endif 922# ret 923# 924 ################################################# 925# Win64 SEH, Structured Exception Handling. 926# 927# Unlike on Unix systems(*) lack of Win64 stack unwinding information 928# has undesired side-effect at run-time: if an exception is raised in 929# assembler subroutine such as those in question (basically we're 930# referring to segmentation violations caused by malformed input 931# parameters), the application is briskly terminated without invoking 932# any exception handlers, most notably without generating memory dump 933# or any user notification whatsoever. This poses a problem. It's 934# possible to address it by registering custom language-specific 935# handler that would restore processor context to the state at 936# subroutine entry point and return "exception is not handled, keep 937# unwinding" code. Writing such handler can be a challenge... But it's 938# doable, though requires certain coding convention. Consider following 939# snippet: 940# 941# .type function,@function 942# function: 943# movq %rsp,%rax # copy rsp to volatile register 944# pushq %r15 # save non-volatile registers 945# pushq %rbx 946# pushq %rbp 947# movq %rsp,%r11 948# subq %rdi,%r11 # prepare [variable] stack frame 949# andq $-64,%r11 950# movq %rax,0(%r11) # check for exceptions 951# movq %r11,%rsp # allocate [variable] stack frame 952# movq %rax,0(%rsp) # save original rsp value 953# magic_point: 954# ... 955# movq 0(%rsp),%rcx # pull original rsp value 956# movq -24(%rcx),%rbp # restore non-volatile registers 957# movq -16(%rcx),%rbx 958# movq -8(%rcx),%r15 959# movq %rcx,%rsp # restore original rsp 960# ret 961# .size function,.-function 962# 963# The key is that up to magic_point copy of original rsp value remains 964# in chosen volatile register and no non-volatile register, except for 965# rsp, is modified. While past magic_point rsp remains constant till 966# the very end of the function. In this case custom language-specific 967# exception handler would look like this: 968# 969# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, 970# CONTEXT *context,DISPATCHER_CONTEXT *disp) 971# { ULONG64 *rsp = (ULONG64 *)context->Rax; 972# if (context->Rip >= magic_point) 973# { rsp = ((ULONG64 **)context->Rsp)[0]; 974# context->Rbp = rsp[-3]; 975# context->Rbx = rsp[-2]; 976# context->R15 = rsp[-1]; 977# } 978# context->Rsp = (ULONG64)rsp; 979# context->Rdi = rsp[1]; 980# context->Rsi = rsp[2]; 981# 982# memcpy (disp->ContextRecord,context,sizeof(CONTEXT)); 983# RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase, 984# dips->ControlPc,disp->FunctionEntry,disp->ContextRecord, 985# &disp->HandlerData,&disp->EstablisherFrame,NULL); 986# return ExceptionContinueSearch; 987# } 988# 989# It's appropriate to implement this handler in assembler, directly in 990# function's module. In order to do that one has to know members' 991# offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant 992# values. Here they are: 993# 994# CONTEXT.Rax 120 995# CONTEXT.Rcx 128 996# CONTEXT.Rdx 136 997# CONTEXT.Rbx 144 998# CONTEXT.Rsp 152 999# CONTEXT.Rbp 160 1000# CONTEXT.Rsi 168 1001# CONTEXT.Rdi 176 1002# CONTEXT.R8 184 1003# CONTEXT.R9 192 1004# CONTEXT.R10 200 1005# CONTEXT.R11 208 1006# CONTEXT.R12 216 1007# CONTEXT.R13 224 1008# CONTEXT.R14 232 1009# CONTEXT.R15 240 1010# CONTEXT.Rip 248 1011# CONTEXT.Xmm6 512 1012# sizeof(CONTEXT) 1232 1013# DISPATCHER_CONTEXT.ControlPc 0 1014# DISPATCHER_CONTEXT.ImageBase 8 1015# DISPATCHER_CONTEXT.FunctionEntry 16 1016# DISPATCHER_CONTEXT.EstablisherFrame 24 1017# DISPATCHER_CONTEXT.TargetIp 32 1018# DISPATCHER_CONTEXT.ContextRecord 40 1019# DISPATCHER_CONTEXT.LanguageHandler 48 1020# DISPATCHER_CONTEXT.HandlerData 56 1021# UNW_FLAG_NHANDLER 0 1022# ExceptionContinueSearch 1 1023# 1024# In order to tie the handler to the function one has to compose 1025# couple of structures: one for .xdata segment and one for .pdata. 1026# 1027# UNWIND_INFO structure for .xdata segment would be 1028# 1029# function_unwind_info: 1030# .byte 9,0,0,0 1031# .rva handler 1032# 1033# This structure designates exception handler for a function with 1034# zero-length prologue, no stack frame or frame register. 1035# 1036# To facilitate composing of .pdata structures, auto-generated "gear" 1037# prologue copies rsp value to rax and denotes next instruction with 1038# .LSEH_begin_{function_name} label. This essentially defines the SEH 1039# styling rule mentioned in the beginning. Position of this label is 1040# chosen in such manner that possible exceptions raised in the "gear" 1041# prologue would be accounted to caller and unwound from latter's frame. 1042# End of function is marked with respective .LSEH_end_{function_name} 1043# label. To summarize, .pdata segment would contain 1044# 1045# .rva .LSEH_begin_function 1046# .rva .LSEH_end_function 1047# .rva function_unwind_info 1048# 1049# Reference to functon_unwind_info from .xdata segment is the anchor. 1050# In case you wonder why references are 32-bit .rvas and not 64-bit 1051# .quads. References put into these two segments are required to be 1052# *relative* to the base address of the current binary module, a.k.a. 1053# image base. No Win64 module, be it .exe or .dll, can be larger than 1054# 2GB and thus such relative references can be and are accommodated in 1055# 32 bits. 1056# 1057# Having reviewed the example function code, one can argue that "movq 1058# %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix 1059# rax would contain an undefined value. If this "offends" you, use 1060# another register and refrain from modifying rax till magic_point is 1061# reached, i.e. as if it was a non-volatile register. If more registers 1062# are required prior [variable] frame setup is completed, note that 1063# nobody says that you can have only one "magic point." You can 1064# "liberate" non-volatile registers by denoting last stack off-load 1065# instruction and reflecting it in finer grade unwind logic in handler. 1066# After all, isn't it why it's called *language-specific* handler... 1067# 1068# Attentive reader can notice that exceptions would be mishandled in 1069# auto-generated "gear" epilogue. Well, exception effectively can't 1070# occur there, because if memory area used by it was subject to 1071# segmentation violation, then it would be raised upon call to the 1072# function (and as already mentioned be accounted to caller, which is 1073# not a problem). If you're still not comfortable, then define tail 1074# "magic point" just prior ret instruction and have handler treat it... 1075# 1076# (*) Note that we're talking about run-time, not debug-time. Lack of 1077# unwind information makes debugging hard on both Windows and 1078# Unix. "Unlike" referes to the fact that on Unix signal handler 1079# will always be invoked, core dumped and appropriate exit code 1080# returned to parent (for user notification). 1081