1 #! /usr/bin/env perl 2 # Copyright 2005-2016 The OpenSSL Project Authors. All Rights Reserved. 3 # 4 # Licensed under the OpenSSL license (the "License"). You may not use 5 # this file except in compliance with the License. You can obtain a copy 6 # in the file LICENSE in the source distribution or at 7 # https://www.openssl.org/source/license.html 8 9 10 # Ascetic x86_64 AT&T to MASM/NASM assembler translator by <appro>. 11 # 12 # Why AT&T to MASM and not vice versa? Several reasons. Because AT&T 13 # format is way easier to parse. Because it's simpler to "gear" from 14 # Unix ABI to Windows one [see cross-reference "card" at the end of 15 # file]. Because Linux targets were available first... 16 # 17 # In addition the script also "distills" code suitable for GNU 18 # assembler, so that it can be compiled with more rigid assemblers, 19 # such as Solaris /usr/ccs/bin/as. 20 # 21 # This translator is not designed to convert *arbitrary* assembler 22 # code from AT&T format to MASM one. It's designed to convert just 23 # enough to provide for dual-ABI OpenSSL modules development... 24 # There *are* limitations and you might have to modify your assembler 25 # code or this script to achieve the desired result... 26 # 27 # Currently recognized limitations: 28 # 29 # - can't use multiple ops per line; 30 # 31 # Dual-ABI styling rules. 32 # 33 # 1. Adhere to Unix register and stack layout [see cross-reference 34 # ABI "card" at the end for explanation]. 35 # 2. Forget about "red zone," stick to more traditional blended 36 # stack frame allocation. If volatile storage is actually required 37 # that is. If not, just leave the stack as is. 38 # 3. Functions tagged with ".type name,@function" get crafted with 39 # unified Win64 prologue and epilogue automatically. If you want 40 # to take care of ABI differences yourself, tag functions as 41 # ".type name,@abi-omnipotent" instead. 42 # 4. To optimize the Win64 prologue you can specify number of input 43 # arguments as ".type name,@function,N." Keep in mind that if N is 44 # larger than 6, then you *have to* write "abi-omnipotent" code, 45 # because >6 cases can't be addressed with unified prologue. 46 # 5. Name local labels as .L*, do *not* use dynamic labels such as 1: 47 # (sorry about latter). 48 # 6. Don't use [or hand-code with .byte] "rep ret." "ret" mnemonic is 49 # required to identify the spots, where to inject Win64 epilogue! 50 # But on the pros, it's then prefixed with rep automatically:-) 51 # 7. Stick to explicit ip-relative addressing. If you have to use 52 # GOTPCREL addressing, stick to mov symbol@GOTPCREL(%rip),%r??. 53 # Both are recognized and translated to proper Win64 addressing 54 # modes. 55 # 56 # 8. In order to provide for structured exception handling unified 57 # Win64 prologue copies %rsp value to %rax. For further details 58 # see SEH paragraph at the end. 59 # 9. .init segment is allowed to contain calls to functions only. 60 # a. If function accepts more than 4 arguments *and* >4th argument 61 # is declared as non 64-bit value, do clear its upper part. 62 64 65 use strict; 66 67 my $flavour = shift; 68 my $output = shift; 69 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; } 70 71 open STDOUT,">$output" || die "can't open $output: $!" 72 if (defined($output)); 73 74 my $gas=1; $gas=0 if ($output =~ /\.asm$/); 75 my $elf=1; $elf=0 if (!$gas); 76 my $win64=0; 77 my $prefix=""; 78 my $decor=".L"; 79 80 my $masmref=8 + 50727*2**-32; # 8.00.50727 shipped with VS2005 81 my $masm=0; 82 my $PTR=" PTR"; 83 84 my $nasmref=2.03; 85 my $nasm=0; 86 87 if ($flavour eq "mingw64") { $gas=1; $elf=0; $win64=1; 88 # TODO(davidben): Before supporting the 89 # mingw64 perlasm flavour, do away with this 90 # environment variable check. 91 die "mingw64 not supported"; 92 $prefix=`echo __USER_LABEL_PREFIX__ | $ENV{CC} -E -P -`; 93 $prefix =~ s|\R$||; # Better chomp 94 } 95 elsif ($flavour eq "macosx") { $gas=1; $elf=0; $prefix="_"; $decor="L\$"; } 96 elsif ($flavour eq "masm") { $gas=0; $elf=0; $masm=$masmref; $win64=1; $decor="\$L\$"; } 97 elsif ($flavour eq "nasm") { $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $decor="\$L\$"; $PTR=""; } 98 elsif (!$gas) { die "unknown flavour $flavour"; } 99 100 my $current_segment; 101 my $current_function; 102 my %globals; 103 104 { package opcode; # pick up opcodes 105 sub re { 106 my ($class, $line) = @_; 107 my $self = {}; 108 my $ret; 109 110 if ($$line =~ /^([a-z][a-z0-9]*)/i) { 111 bless $self,$class; 112 $self->{op} = $1; 113 $ret = $self; 114 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 115 116 undef $self->{sz}; 117 if ($self->{op} =~ /^(movz)x?([bw]).*/) { # movz is pain... 118 $self->{op} = $1; 119 $self->{sz} = $2; 120 } elsif ($self->{op} =~ /call|jmp/) { 121 $self->{sz} = ""; 122 } elsif ($self->{op} =~ /^p/ && $' !~ /^(ush|op|insrw)/) { # SSEn 123 $self->{sz} = ""; 124 } elsif ($self->{op} =~ /^[vk]/) { # VEX or k* such as kmov 125 $self->{sz} = ""; 126 } elsif ($self->{op} =~ /mov[dq]/ && $$line =~ /%xmm/) { 127 $self->{sz} = ""; 128 } elsif ($self->{op} =~ /^or([qlwb])$/) { 129 $self->{op} = "or"; 130 $self->{sz} = $1; 131 } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) { 132 $self->{op} = $1; 133 $self->{sz} = $2; 134 } 135 } 136 $ret; 137 } 138 sub size { 139 my ($self, $sz) = @_; 140 $self->{sz} = $sz if (defined($sz) && !defined($self->{sz})); 141 $self->{sz}; 142 } 143 sub out { 144 my $self = shift; 145 if ($gas) { 146 if ($self->{op} eq "movz") { # movz is pain... 147 sprintf "%s%s%s",$self->{op},$self->{sz},shift; 148 } elsif ($self->{op} =~ /^set/) { 149 "$self->{op}"; 150 } elsif ($self->{op} eq "ret") { 151 my $epilogue = ""; 152 if ($win64 && $current_function->{abi} eq "svr4") { 153 $epilogue = "movq 8(%rsp),%rdi\n\t" . 154 "movq 16(%rsp),%rsi\n\t"; 155 } 156 $epilogue . ".byte 0xf3,0xc3"; 157 } elsif ($self->{op} eq "call" && !$elf && $current_segment eq ".init") { 158 ".p2align\t3\n\t.quad"; 159 } else { 160 "$self->{op}$self->{sz}"; 161 } 162 } else { 163 $self->{op} =~ s/^movz/movzx/; 164 if ($self->{op} eq "ret") { 165 $self->{op} = ""; 166 if ($win64 && $current_function->{abi} eq "svr4") { 167 $self->{op} = "mov rdi,QWORD$PTR\[8+rsp\]\t;WIN64 epilogue\n\t". 168 "mov rsi,QWORD$PTR\[16+rsp\]\n\t"; 169 } 170 $self->{op} .= "DB\t0F3h,0C3h\t\t;repret"; 171 } elsif ($self->{op} =~ /^(pop|push)f/) { 172 $self->{op} .= $self->{sz}; 173 } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") { 174 $self->{op} = "\tDQ"; 175 } 176 $self->{op}; 177 } 178 } 179 sub mnemonic { 180 my ($self, $op) = @_; 181 $self->{op}=$op if (defined($op)); 182 $self->{op}; 183 } 184 } 185 { package const; # pick up constants, which start with $ 186 sub re { 187 my ($class, $line) = @_; 188 my $self = {}; 189 my $ret; 190 191 if ($$line =~ /^\$([^,]+)/) { 192 bless $self, $class; 193 $self->{value} = $1; 194 $ret = $self; 195 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 196 } 197 $ret; 198 } 199 sub out { 200 my $self = shift; 201 202 $self->{value} =~ s/\b(0b[0-1]+)/oct($1)/eig; 203 if ($gas) { 204 # Solaris /usr/ccs/bin/as can't handle multiplications 205 # in $self->{value} 206 my $value = $self->{value}; 207 no warnings; # oct might complain about overflow, ignore here... 208 $value =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi; 209 if ($value =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg) { 210 $self->{value} = $value; 211 } 212 sprintf "\$%s",$self->{value}; 213 } else { 214 my $value = $self->{value}; 215 $value =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm); 216 sprintf "%s",$value; 217 } 218 } 219 } 220 { package ea; # pick up effective addresses: expr(%reg,%reg,scale) 221 222 my %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR", 223 l=>"DWORD$PTR", d=>"DWORD$PTR", 224 q=>"QWORD$PTR", o=>"OWORD$PTR", 225 x=>"XMMWORD$PTR", y=>"YMMWORD$PTR", 226 z=>"ZMMWORD$PTR" ) if (!$gas); 227 228 sub re { 229 my ($class, $line, $opcode) = @_; 230 my $self = {}; 231 my $ret; 232 233 # optional * ----vvv--- appears in indirect jmp/call 234 if ($$line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)((?:{[^}]+})*)/) { 235 bless $self, $class; 236 $self->{asterisk} = $1; 237 $self->{label} = $2; 238 ($self->{base},$self->{index},$self->{scale})=split(/,/,$3); 239 $self->{scale} = 1 if (!defined($self->{scale})); 240 $self->{opmask} = $4; 241 $ret = $self; 242 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 243 244 if ($win64 && $self->{label} =~ s/\@GOTPCREL//) { 245 die if ($opcode->mnemonic() ne "mov"); 246 $opcode->mnemonic("lea"); 247 } 248 $self->{base} =~ s/^%//; 249 $self->{index} =~ s/^%// if (defined($self->{index})); 250 $self->{opcode} = $opcode; 251 } 252 $ret; 253 } 254 sub size {} 255 sub out { 256 my ($self, $sz) = @_; 257 258 $self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 259 $self->{label} =~ s/\.L/$decor/g; 260 261 # Silently convert all EAs to 64-bit. This is required for 262 # elder GNU assembler and results in more compact code, 263 # *but* most importantly AES module depends on this feature! 264 $self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; 265 $self->{base} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; 266 267 # Solaris /usr/ccs/bin/as can't handle multiplications 268 # in $self->{label}... 269 use integer; 270 $self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi; 271 $self->{label} =~ s/\b([0-9]+\s*[\*\/\%]\s*[0-9]+)\b/eval($1)/eg; 272 273 # Some assemblers insist on signed presentation of 32-bit 274 # offsets, but sign extension is a tricky business in perl... 275 if ((1<<31)<<1) { 276 $self->{label} =~ s/\b([0-9]+)\b/$1<<32>>32/eg; 277 } else { 278 $self->{label} =~ s/\b([0-9]+)\b/$1>>0/eg; 279 } 280 281 # if base register is %rbp or %r13, see if it's possible to 282 # flip base and index registers [for better performance] 283 if (!$self->{label} && $self->{index} && $self->{scale}==1 && 284 $self->{base} =~ /(rbp|r13)/) { 285 $self->{base} = $self->{index}; $self->{index} = $1; 286 } 287 288 if ($gas) { 289 $self->{label} =~ s/^___imp_/__imp__/ if ($flavour eq "mingw64"); 290 291 if (defined($self->{index})) { 292 sprintf "%s%s(%s,%%%s,%d)%s", 293 $self->{asterisk},$self->{label}, 294 $self->{base}?"%$self->{base}":"", 295 $self->{index},$self->{scale}, 296 $self->{opmask}; 297 } else { 298 sprintf "%s%s(%%%s)%s", $self->{asterisk},$self->{label}, 299 $self->{base},$self->{opmask}; 300 } 301 } else { 302 $self->{label} =~ s/\./\$/g; 303 $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig; 304 $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/); 305 306 my $mnemonic = $self->{opcode}->mnemonic(); 307 ($self->{asterisk}) && ($sz="q") || 308 ($mnemonic =~ /^v?mov([qd])$/) && ($sz=$1) || 309 ($mnemonic =~ /^v?pinsr([qdwb])$/) && ($sz=$1) || 310 ($mnemonic =~ /^vpbroadcast([qdwb])$/) && ($sz=$1) || 311 ($mnemonic =~ /^v(?!perm)[a-z]+[fi]128$/) && ($sz="x"); 312 313 $self->{opmask} =~ s/%(k[0-7])/$1/; 314 315 if (defined($self->{index})) { 316 sprintf "%s[%s%s*%d%s]%s",$szmap{$sz}, 317 $self->{label}?"$self->{label}+":"", 318 $self->{index},$self->{scale}, 319 $self->{base}?"+$self->{base}":"", 320 $self->{opmask}; 321 } elsif ($self->{base} eq "rip") { 322 sprintf "%s[%s]",$szmap{$sz},$self->{label}; 323 } else { 324 sprintf "%s[%s%s]%s", $szmap{$sz}, 325 $self->{label}?"$self->{label}+":"", 326 $self->{base},$self->{opmask}; 327 } 328 } 329 } 330 } 331 { package register; # pick up registers, which start with %. 332 sub re { 333 my ($class, $line, $opcode) = @_; 334 my $self = {}; 335 my $ret; 336 337 # optional * ----vvv--- appears in indirect jmp/call 338 if ($$line =~ /^(\*?)%(\w+)((?:{[^}]+})*)/) { 339 bless $self,$class; 340 $self->{asterisk} = $1; 341 $self->{value} = $2; 342 $self->{opmask} = $3; 343 $opcode->size($self->size()); 344 $ret = $self; 345 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 346 } 347 $ret; 348 } 349 sub size { 350 my $self = shift; 351 my $ret; 352 353 if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; } 354 elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; } 355 elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; } 356 elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; } 357 elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; } 358 elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; } 359 elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; } 360 elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; } 361 362 $ret; 363 } 364 sub out { 365 my $self = shift; 366 if ($gas) { sprintf "%s%%%s%s", $self->{asterisk}, 367 $self->{value}, 368 $self->{opmask}; } 369 else { $self->{opmask} =~ s/%(k[0-7])/$1/; 370 $self->{value}.$self->{opmask}; } 371 } 372 } 373 { package label; # pick up labels, which end with : 374 sub re { 375 my ($class, $line) = @_; 376 my $self = {}; 377 my $ret; 378 379 if ($$line =~ /(^[\.\w]+)\:/) { 380 bless $self,$class; 381 $self->{value} = $1; 382 $ret = $self; 383 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 384 385 $self->{value} =~ s/^\.L/$decor/; 386 } 387 $ret; 388 } 389 sub out { 390 my $self = shift; 391 392 if ($gas) { 393 my $func = ($globals{$self->{value}} or $self->{value}) . ":"; 394 if ($win64 && $current_function->{name} eq $self->{value} 395 && $current_function->{abi} eq "svr4") { 396 $func .= "\n"; 397 $func .= " movq %rdi,8(%rsp)\n"; 398 $func .= " movq %rsi,16(%rsp)\n"; 399 $func .= " movq %rsp,%rax\n"; 400 $func .= "${decor}SEH_begin_$current_function->{name}:\n"; 401 my $narg = $current_function->{narg}; 402 $narg=6 if (!defined($narg)); 403 $func .= " movq %rcx,%rdi\n" if ($narg>0); 404 $func .= " movq %rdx,%rsi\n" if ($narg>1); 405 $func .= " movq %r8,%rdx\n" if ($narg>2); 406 $func .= " movq %r9,%rcx\n" if ($narg>3); 407 $func .= " movq 40(%rsp),%r8\n" if ($narg>4); 408 $func .= " movq 48(%rsp),%r9\n" if ($narg>5); 409 } 410 $func; 411 } elsif ($self->{value} ne "$current_function->{name}") { 412 # Make all labels in masm global. 413 $self->{value} .= ":" if ($masm); 414 $self->{value} . ":"; 415 } elsif ($win64 && $current_function->{abi} eq "svr4") { 416 my $func = "$current_function->{name}" . 417 ($nasm ? ":" : "\tPROC $current_function->{scope}") . 418 "\n"; 419 $func .= " mov QWORD$PTR\[8+rsp\],rdi\t;WIN64 prologue\n"; 420 $func .= " mov QWORD$PTR\[16+rsp\],rsi\n"; 421 $func .= " mov rax,rsp\n"; 422 $func .= "${decor}SEH_begin_$current_function->{name}:"; 423 $func .= ":" if ($masm); 424 $func .= "\n"; 425 my $narg = $current_function->{narg}; 426 $narg=6 if (!defined($narg)); 427 $func .= " mov rdi,rcx\n" if ($narg>0); 428 $func .= " mov rsi,rdx\n" if ($narg>1); 429 $func .= " mov rdx,r8\n" if ($narg>2); 430 $func .= " mov rcx,r9\n" if ($narg>3); 431 $func .= " mov r8,QWORD$PTR\[40+rsp\]\n" if ($narg>4); 432 $func .= " mov r9,QWORD$PTR\[48+rsp\]\n" if ($narg>5); 433 $func .= "\n"; 434 } else { 435 "$current_function->{name}". 436 ($nasm ? ":" : "\tPROC $current_function->{scope}"); 437 } 438 } 439 } 440 { package expr; # pick up expressions 441 sub re { 442 my ($class, $line, $opcode) = @_; 443 my $self = {}; 444 my $ret; 445 446 if ($$line =~ /(^[^,]+)/) { 447 bless $self,$class; 448 $self->{value} = $1; 449 $ret = $self; 450 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 451 452 $self->{value} =~ s/\@PLT// if (!$elf); 453 $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 454 $self->{value} =~ s/\.L/$decor/g; 455 $self->{opcode} = $opcode; 456 } 457 $ret; 458 } 459 sub out { 460 my $self = shift; 461 if ($nasm && $self->{opcode}->mnemonic()=~m/^j(?![re]cxz)/) { 462 "NEAR ".$self->{value}; 463 } else { 464 $self->{value}; 465 } 466 } 467 } 468 { package cfi_directive; 469 # CFI directives annotate instructions that are significant for 470 # stack unwinding procedure compliant with DWARF specification, 471 # see http://dwarfstd.org/. Besides naturally expected for this 472 # script platform-specific filtering function, this module adds 473 # three auxiliary synthetic directives not recognized by [GNU] 474 # assembler: 475 # 476 # - .cfi_push to annotate push instructions in prologue, which 477 # translates to .cfi_adjust_cfa_offset (if needed) and 478 # .cfi_offset; 479 # - .cfi_pop to annotate pop instructions in epilogue, which 480 # translates to .cfi_adjust_cfa_offset (if needed) and 481 # .cfi_restore; 482 # - [and most notably] .cfi_cfa_expression which encodes 483 # DW_CFA_def_cfa_expression and passes it to .cfi_escape as 484 # byte vector; 485 # 486 # CFA expressions were introduced in DWARF specification version 487 # 3 and describe how to deduce CFA, Canonical Frame Address. This 488 # becomes handy if your stack frame is variable and you can't 489 # spare register for [previous] frame pointer. Suggested directive 490 # syntax is made-up mix of DWARF operator suffixes [subset of] 491 # and references to registers with optional bias. Following example 492 # describes offloaded *original* stack pointer at specific offset 493 # from *current* stack pointer: 494 # 495 # .cfi_cfa_expression %rsp+40,deref,+8 496 # 497 # Final +8 has everything to do with the fact that CFA is defined 498 # as reference to top of caller's stack, and on x86_64 call to 499 # subroutine pushes 8-byte return address. In other words original 500 # stack pointer upon entry to a subroutine is 8 bytes off from CFA. 501 502 # Below constants are taken from "DWARF Expressions" section of the 503 # DWARF specification, section is numbered 7.7 in versions 3 and 4. 504 my %DW_OP_simple = ( # no-arg operators, mapped directly 505 deref => 0x06, dup => 0x12, 506 drop => 0x13, over => 0x14, 507 pick => 0x15, swap => 0x16, 508 rot => 0x17, xderef => 0x18, 509 510 abs => 0x19, and => 0x1a, 511 div => 0x1b, minus => 0x1c, 512 mod => 0x1d, mul => 0x1e, 513 neg => 0x1f, not => 0x20, 514 or => 0x21, plus => 0x22, 515 shl => 0x24, shr => 0x25, 516 shra => 0x26, xor => 0x27, 517 ); 518 519 my %DW_OP_complex = ( # used in specific subroutines 520 constu => 0x10, # uleb128 521 consts => 0x11, # sleb128 522 plus_uconst => 0x23, # uleb128 523 lit0 => 0x30, # add 0-31 to opcode 524 reg0 => 0x50, # add 0-31 to opcode 525 breg0 => 0x70, # add 0-31 to opcole, sleb128 526 regx => 0x90, # uleb28 527 fbreg => 0x91, # sleb128 528 bregx => 0x92, # uleb128, sleb128 529 piece => 0x93, # uleb128 530 ); 531 532 # Following constants are defined in x86_64 ABI supplement, for 533 # example available at https://www.uclibc.org/docs/psABI-x86_64.pdf, 534 # see section 3.7 "Stack Unwind Algorithm". 535 my %DW_reg_idx = ( 536 "%rax"=>0, "%rdx"=>1, "%rcx"=>2, "%rbx"=>3, 537 "%rsi"=>4, "%rdi"=>5, "%rbp"=>6, "%rsp"=>7, 538 "%r8" =>8, "%r9" =>9, "%r10"=>10, "%r11"=>11, 539 "%r12"=>12, "%r13"=>13, "%r14"=>14, "%r15"=>15 540 ); 541 542 my ($cfa_reg, $cfa_rsp); 543 my @cfa_stack; 544 545 # [us]leb128 format is variable-length integer representation base 546 # 2^128, with most significant bit of each byte being 0 denoting 547 # *last* most significant digit. See "Variable Length Data" in the 548 # DWARF specification, numbered 7.6 at least in versions 3 and 4. 549 sub sleb128 { 550 use integer; # get right shift extend sign 551 552 my $val = shift; 553 my $sign = ($val < 0) ? -1 : 0; 554 my @ret = (); 555 556 while(1) { 557 push @ret, $val&0x7f; 558 559 # see if remaining bits are same and equal to most 560 # significant bit of the current digit, if so, it's 561 # last digit... 562 last if (($val>>6) == $sign); 563 564 @ret[-1] |= 0x80; 565 $val >>= 7; 566 } 567 568 return @ret; 569 } 570 sub uleb128 { 571 my $val = shift; 572 my @ret = (); 573 574 while(1) { 575 push @ret, $val&0x7f; 576 577 # see if it's last significant digit... 578 last if (($val >>= 7) == 0); 579 580 @ret[-1] |= 0x80; 581 } 582 583 return @ret; 584 } 585 sub const { 586 my $val = shift; 587 588 if ($val >= 0 && $val < 32) { 589 return ($DW_OP_complex{lit0}+$val); 590 } 591 return ($DW_OP_complex{consts}, sleb128($val)); 592 } 593 sub reg { 594 my $val = shift; 595 596 return if ($val !~ m/^(%r\w+)(?:([\+\-])((?:0x)?[0-9a-f]+))?/); 597 598 my $reg = $DW_reg_idx{$1}; 599 my $off = eval ("0 $2 $3"); 600 601 return (($DW_OP_complex{breg0} + $reg), sleb128($off)); 602 # Yes, we use DW_OP_bregX+0 to push register value and not 603 # DW_OP_regX, because latter would require even DW_OP_piece, 604 # which would be a waste under the circumstances. If you have 605 # to use DWP_OP_reg, use "regx:N"... 606 } 607 sub cfa_expression { 608 my $line = shift; 609 my @ret; 610 611 foreach my $token (split(/,\s*/,$line)) { 612 if ($token =~ /^%r/) { 613 push @ret,reg($token); 614 } elsif ($token =~ /((?:0x)?[0-9a-f]+)\((%r\w+)\)/) { 615 push @ret,reg("$2+$1"); 616 } elsif ($token =~ /(\w+):(\-?(?:0x)?[0-9a-f]+)(U?)/i) { 617 my $i = 1*eval($2); 618 push @ret,$DW_OP_complex{$1}, ($3 ? uleb128($i) : sleb128($i)); 619 } elsif (my $i = 1*eval($token) or $token eq "0") { 620 if ($token =~ /^\+/) { 621 push @ret,$DW_OP_complex{plus_uconst},uleb128($i); 622 } else { 623 push @ret,const($i); 624 } 625 } else { 626 push @ret,$DW_OP_simple{$token}; 627 } 628 } 629 630 # Finally we return DW_CFA_def_cfa_expression, 15, followed by 631 # length of the expression and of course the expression itself. 632 return (15,scalar(@ret),@ret); 633 } 634 sub re { 635 my ($class, $line) = @_; 636 my $self = {}; 637 my $ret; 638 639 if ($$line =~ s/^\s*\.cfi_(\w+)\s*//) { 640 bless $self,$class; 641 $ret = $self; 642 undef $self->{value}; 643 my $dir = $1; 644 645 SWITCH: for ($dir) { 646 # What is $cfa_rsp? Effectively it's difference between %rsp 647 # value and current CFA, Canonical Frame Address, which is 648 # why it starts with -8. Recall that CFA is top of caller's 649 # stack... 650 /startproc/ && do { ($cfa_reg, $cfa_rsp) = ("%rsp", -8); last; }; 651 /endproc/ && do { ($cfa_reg, $cfa_rsp) = ("%rsp", 0); last; }; 652 /def_cfa_register/ 653 && do { $cfa_reg = $$line; last; }; 654 /def_cfa_offset/ 655 && do { $cfa_rsp = -1*eval($$line) if ($cfa_reg eq "%rsp"); 656 last; 657 }; 658 /adjust_cfa_offset/ 659 && do { $cfa_rsp -= 1*eval($$line) if ($cfa_reg eq "%rsp"); 660 last; 661 }; 662 /def_cfa/ && do { if ($$line =~ /(%r\w+)\s*,\s*(.+)/) { 663 $cfa_reg = $1; 664 $cfa_rsp = -1*eval($2) if ($cfa_reg eq "%rsp"); 665 } 666 last; 667 }; 668 /push/ && do { $dir = undef; 669 $cfa_rsp -= 8; 670 if ($cfa_reg eq "%rsp") { 671 $self->{value} = ".cfi_adjust_cfa_offset\t8\n"; 672 } 673 $self->{value} .= ".cfi_offset\t$$line,$cfa_rsp"; 674 last; 675 }; 676 /pop/ && do { $dir = undef; 677 $cfa_rsp += 8; 678 if ($cfa_reg eq "%rsp") { 679 $self->{value} = ".cfi_adjust_cfa_offset\t-8\n"; 680 } 681 $self->{value} .= ".cfi_restore\t$$line"; 682 last; 683 }; 684 /cfa_expression/ 685 && do { $dir = undef; 686 $self->{value} = ".cfi_escape\t" . 687 join(",", map(sprintf("0x%02x", $_), 688 cfa_expression($$line))); 689 last; 690 }; 691 /remember_state/ 692 && do { push @cfa_stack, [$cfa_reg, $cfa_rsp]; 693 last; 694 }; 695 /restore_state/ 696 && do { ($cfa_reg, $cfa_rsp) = @{pop @cfa_stack}; 697 last; 698 }; 699 } 700 701 $self->{value} = ".cfi_$dir\t$$line" if ($dir); 702 703 $$line = ""; 704 } 705 706 return $ret; 707 } 708 sub out { 709 my $self = shift; 710 return ($elf ? $self->{value} : undef); 711 } 712 } 713 { package directive; # pick up directives, which start with . 714 sub re { 715 my ($class, $line) = @_; 716 my $self = {}; 717 my $ret; 718 my $dir; 719 720 # chain-call to cfi_directive 721 $ret = cfi_directive->re($line) and return $ret; 722 723 if ($$line =~ /^\s*(\.\w+)/) { 724 bless $self,$class; 725 $dir = $1; 726 $ret = $self; 727 undef $self->{value}; 728 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 729 730 SWITCH: for ($dir) { 731 /\.global|\.globl|\.extern/ 732 && do { $globals{$$line} = $prefix . $$line; 733 $$line = $globals{$$line} if ($prefix); 734 last; 735 }; 736 /\.type/ && do { my ($sym,$type,$narg) = split(/\s*,\s*/,$$line); 737 if ($type eq "\@function") { 738 undef $current_function; 739 $current_function->{name} = $sym; 740 $current_function->{abi} = "svr4"; 741 $current_function->{narg} = $narg; 742 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 743 } elsif ($type eq "\@abi-omnipotent") { 744 undef $current_function; 745 $current_function->{name} = $sym; 746 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 747 } 748 $$line =~ s/\@abi\-omnipotent/\@function/; 749 $$line =~ s/\@function.*/\@function/; 750 last; 751 }; 752 /\.asciz/ && do { if ($$line =~ /^"(.*)"$/) { 753 $dir = ".byte"; 754 $$line = join(",",unpack("C*",$1),0); 755 } 756 last; 757 }; 758 /\.rva|\.long|\.quad|\.byte/ 759 && do { $$line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 760 $$line =~ s/\.L/$decor/g; 761 last; 762 }; 763 } 764 765 if ($gas) { 766 $self->{value} = $dir . "\t" . $$line; 767 768 if ($dir =~ /\.extern/) { 769 if ($flavour eq "elf") { 770 $self->{value} .= "\n.hidden $$line"; 771 } else { 772 $self->{value} = ""; 773 } 774 } elsif (!$elf && $dir =~ /\.type/) { 775 $self->{value} = ""; 776 $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" . 777 (defined($globals{$1})?".scl 2;":".scl 3;") . 778 "\t.type 32;\t.endef" 779 if ($win64 && $$line =~ /([^,]+),\@function/); 780 } elsif (!$elf && $dir =~ /\.size/) { 781 $self->{value} = ""; 782 if (defined($current_function)) { 783 $self->{value} .= "${decor}SEH_end_$current_function->{name}:" 784 if ($win64 && $current_function->{abi} eq "svr4"); 785 undef $current_function; 786 } 787 } elsif (!$elf && $dir =~ /\.align/) { 788 $self->{value} = ".p2align\t" . (log($$line)/log(2)); 789 } elsif ($dir eq ".section") { 790 $current_segment=$$line; 791 if (!$elf && $current_segment eq ".init") { 792 if ($flavour eq "macosx") { $self->{value} = ".mod_init_func"; } 793 elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.ctors"; } 794 } 795 } elsif ($dir =~ /\.(text|data)/) { 796 $current_segment=".$1"; 797 } elsif ($dir =~ /\.global|\.globl|\.extern/) { 798 if ($flavour eq "macosx") { 799 $self->{value} .= "\n.private_extern $$line"; 800 } else { 801 $self->{value} .= "\n.hidden $$line"; 802 } 803 } elsif ($dir =~ /\.hidden/) { 804 if ($flavour eq "macosx") { $self->{value} = ".private_extern\t$prefix$$line"; } 805 elsif ($flavour eq "mingw64") { $self->{value} = ""; } 806 } elsif ($dir =~ /\.comm/) { 807 $self->{value} = "$dir\t$prefix$$line"; 808 $self->{value} =~ s|,([0-9]+),([0-9]+)$|",$1,".log($2)/log(2)|e if ($flavour eq "macosx"); 809 } 810 $$line = ""; 811 return $self; 812 } 813 814 # non-gas case or nasm/masm 815 SWITCH: for ($dir) { 816 /\.text/ && do { my $v=undef; 817 if ($nasm) { 818 $v="section .text code align=64\n"; 819 } else { 820 $v="$current_segment\tENDS\n" if ($current_segment); 821 $current_segment = ".text\$"; 822 $v.="$current_segment\tSEGMENT "; 823 $v.=$masm>=$masmref ? "ALIGN(256)" : "PAGE"; 824 $v.=" 'CODE'"; 825 } 826 $self->{value} = $v; 827 last; 828 }; 829 /\.data/ && do { my $v=undef; 830 if ($nasm) { 831 $v="section .data data align=8\n"; 832 } else { 833 $v="$current_segment\tENDS\n" if ($current_segment); 834 $current_segment = "_DATA"; 835 $v.="$current_segment\tSEGMENT"; 836 } 837 $self->{value} = $v; 838 last; 839 }; 840 /\.section/ && do { my $v=undef; 841 $$line =~ s/([^,]*).*/$1/; 842 $$line = ".CRT\$XCU" if ($$line eq ".init"); 843 if ($nasm) { 844 $v="section $$line"; 845 if ($$line=~/\.([px])data/) { 846 $v.=" rdata align="; 847 $v.=$1 eq "p"? 4 : 8; 848 } elsif ($$line=~/\.CRT\$/i) { 849 $v.=" rdata align=8"; 850 } 851 } else { 852 $v="$current_segment\tENDS\n" if ($current_segment); 853 $v.="$$line\tSEGMENT"; 854 if ($$line=~/\.([px])data/) { 855 $v.=" READONLY"; 856 $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref); 857 } elsif ($$line=~/\.CRT\$/i) { 858 $v.=" READONLY "; 859 $v.=$masm>=$masmref ? "ALIGN(8)" : "DWORD"; 860 } 861 } 862 $current_segment = $$line; 863 $self->{value} = $v; 864 last; 865 }; 866 /\.extern/ && do { $self->{value} = "EXTERN\t".$$line; 867 $self->{value} .= ":NEAR" if ($masm); 868 last; 869 }; 870 /\.globl|.global/ 871 && do { $self->{value} = $masm?"PUBLIC":"global"; 872 $self->{value} .= "\t".$$line; 873 last; 874 }; 875 /\.size/ && do { if (defined($current_function)) { 876 undef $self->{value}; 877 if ($current_function->{abi} eq "svr4") { 878 $self->{value}="${decor}SEH_end_$current_function->{name}:"; 879 $self->{value}.=":\n" if($masm); 880 } 881 $self->{value}.="$current_function->{name}\tENDP" if($masm && $current_function->{name}); 882 undef $current_function; 883 } 884 last; 885 }; 886 /\.align/ && do { my $max = ($masm && $masm>=$masmref) ? 256 : 4096; 887 $self->{value} = "ALIGN\t".($$line>$max?$max:$$line); 888 last; 889 }; 890 /\.(value|long|rva|quad)/ 891 && do { my $sz = substr($1,0,1); 892 my @arr = split(/,\s*/,$$line); 893 my $last = pop(@arr); 894 my $conv = sub { my $var=shift; 895 $var=~s/^(0b[0-1]+)/oct($1)/eig; 896 $var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm); 897 if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva")) 898 { $var=~s/^([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; } 899 $var; 900 }; 901 902 $sz =~ tr/bvlrq/BWDDQ/; 903 $self->{value} = "\tD$sz\t"; 904 for (@arr) { $self->{value} .= &$conv($_).","; } 905 $self->{value} .= &$conv($last); 906 last; 907 }; 908 /\.byte/ && do { my @str=split(/,\s*/,$$line); 909 map(s/(0b[0-1]+)/oct($1)/eig,@str); 910 map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm); 911 while ($#str>15) { 912 $self->{value}.="DB\t" 913 .join(",",@str[0..15])."\n"; 914 foreach (0..15) { shift @str; } 915 } 916 $self->{value}.="DB\t" 917 .join(",",@str) if (@str); 918 last; 919 }; 920 /\.comm/ && do { my @str=split(/,\s*/,$$line); 921 my $v=undef; 922 if ($nasm) { 923 $v.="common $prefix@str[0] @str[1]"; 924 } else { 925 $v="$current_segment\tENDS\n" if ($current_segment); 926 $current_segment = "_DATA"; 927 $v.="$current_segment\tSEGMENT\n"; 928 $v.="COMM @str[0]:DWORD:".@str[1]/4; 929 } 930 $self->{value} = $v; 931 last; 932 }; 933 } 934 $$line = ""; 935 } 936 937 $ret; 938 } 939 sub out { 940 my $self = shift; 941 $self->{value}; 942 } 943 } 944 945 # Upon initial x86_64 introduction SSE>2 extensions were not introduced 946 # yet. In order not to be bothered by tracing exact assembler versions, 947 # but at the same time to provide a bare security minimum of AES-NI, we 948 # hard-code some instructions. Extensions past AES-NI on the other hand 949 # are traced by examining assembler version in individual perlasm 950 # modules... 951 952 my %regrm = ( "%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3, 953 "%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7 ); 954 955 sub rex { 956 my $opcode=shift; 957 my ($dst,$src,$rex)=@_; 958 959 $rex|=0x04 if($dst>=8); 960 $rex|=0x01 if($src>=8); 961 push @$opcode,($rex|0x40) if ($rex); 962 } 963 964 my $movq = sub { # elderly gas can't handle inter-register movq 965 my $arg = shift; 966 my @opcode=(0x66); 967 if ($arg =~ /%xmm([0-9]+),\s*%r(\w+)/) { 968 my ($src,$dst)=($1,$2); 969 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 970 rex(\@opcode,$src,$dst,0x8); 971 push @opcode,0x0f,0x7e; 972 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 973 @opcode; 974 } elsif ($arg =~ /%r(\w+),\s*%xmm([0-9]+)/) { 975 my ($src,$dst)=($2,$1); 976 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 977 rex(\@opcode,$src,$dst,0x8); 978 push @opcode,0x0f,0x6e; 979 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 980 @opcode; 981 } else { 982 (); 983 } 984 }; 985 986 my $pextrd = sub { 987 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*(%\w+)/) { 988 my @opcode=(0x66); 989 my $imm=$1; 990 my $src=$2; 991 my $dst=$3; 992 if ($dst =~ /%r([0-9]+)d/) { $dst = $1; } 993 elsif ($dst =~ /%e/) { $dst = $regrm{$dst}; } 994 rex(\@opcode,$src,$dst); 995 push @opcode,0x0f,0x3a,0x16; 996 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 997 push @opcode,$imm; 998 @opcode; 999 } else { 1000 (); 1001 } 1002 }; 1003 1004 my $pinsrd = sub { 1005 if (shift =~ /\$([0-9]+),\s*(%\w+),\s*%xmm([0-9]+)/) { 1006 my @opcode=(0x66); 1007 my $imm=$1; 1008 my $src=$2; 1009 my $dst=$3; 1010 if ($src =~ /%r([0-9]+)/) { $src = $1; } 1011 elsif ($src =~ /%e/) { $src = $regrm{$src}; } 1012 rex(\@opcode,$dst,$src); 1013 push @opcode,0x0f,0x3a,0x22; 1014 push @opcode,0xc0|(($dst&7)<<3)|($src&7); # ModR/M 1015 push @opcode,$imm; 1016 @opcode; 1017 } else { 1018 (); 1019 } 1020 }; 1021 1022 my $pshufb = sub { 1023 if (shift =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) { 1024 my @opcode=(0x66); 1025 rex(\@opcode,$2,$1); 1026 push @opcode,0x0f,0x38,0x00; 1027 push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M 1028 @opcode; 1029 } else { 1030 (); 1031 } 1032 }; 1033 1034 my $palignr = sub { 1035 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 1036 my @opcode=(0x66); 1037 rex(\@opcode,$3,$2); 1038 push @opcode,0x0f,0x3a,0x0f; 1039 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 1040 push @opcode,$1; 1041 @opcode; 1042 } else { 1043 (); 1044 } 1045 }; 1046 1047 my $pclmulqdq = sub { 1048 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 1049 my @opcode=(0x66); 1050 rex(\@opcode,$3,$2); 1051 push @opcode,0x0f,0x3a,0x44; 1052 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 1053 my $c=$1; 1054 push @opcode,$c=~/^0/?oct($c):$c; 1055 @opcode; 1056 } else { 1057 (); 1058 } 1059 }; 1060 1061 my $rdrand = sub { 1062 if (shift =~ /%[er](\w+)/) { 1063 my @opcode=(); 1064 my $dst=$1; 1065 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 1066 rex(\@opcode,0,$dst,8); 1067 push @opcode,0x0f,0xc7,0xf0|($dst&7); 1068 @opcode; 1069 } else { 1070 (); 1071 } 1072 }; 1073 1074 my $rdseed = sub { 1075 if (shift =~ /%[er](\w+)/) { 1076 my @opcode=(); 1077 my $dst=$1; 1078 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 1079 rex(\@opcode,0,$dst,8); 1080 push @opcode,0x0f,0xc7,0xf8|($dst&7); 1081 @opcode; 1082 } else { 1083 (); 1084 } 1085 }; 1086 1087 # Not all AVX-capable assemblers recognize AMD XOP extension. Since we 1088 # are using only two instructions hand-code them in order to be excused 1089 # from chasing assembler versions... 1090 1091 sub rxb { 1092 my $opcode=shift; 1093 my ($dst,$src1,$src2,$rxb)=@_; 1094 1095 $rxb|=0x7<<5; 1096 $rxb&=~(0x04<<5) if($dst>=8); 1097 $rxb&=~(0x01<<5) if($src1>=8); 1098 $rxb&=~(0x02<<5) if($src2>=8); 1099 push @$opcode,$rxb; 1100 } 1101 1102 my $vprotd = sub { 1103 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 1104 my @opcode=(0x8f); 1105 rxb(\@opcode,$3,$2,-1,0x08); 1106 push @opcode,0x78,0xc2; 1107 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 1108 my $c=$1; 1109 push @opcode,$c=~/^0/?oct($c):$c; 1110 @opcode; 1111 } else { 1112 (); 1113 } 1114 }; 1115 1116 my $vprotq = sub { 1117 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 1118 my @opcode=(0x8f); 1119 rxb(\@opcode,$3,$2,-1,0x08); 1120 push @opcode,0x78,0xc3; 1121 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 1122 my $c=$1; 1123 push @opcode,$c=~/^0/?oct($c):$c; 1124 @opcode; 1125 } else { 1126 (); 1127 } 1128 }; 1129 1130 # Intel Control-flow Enforcement Technology extension. All functions and 1131 # indirect branch targets will have to start with this instruction... 1132 1133 my $endbranch = sub { 1134 (0xf3,0x0f,0x1e,0xfa); 1135 }; 1136 1137 ######################################################################## 1138 1139 { 1140 my $comment = "#"; 1141 $comment = ";" if ($masm || $nasm); 1142 print <<___; 1143 $comment This file is generated from a similarly-named Perl script in the BoringSSL 1144 $comment source tree. Do not edit by hand. 1145 1146 ___ 1147 } 1148 1149 if ($nasm) { 1150 print <<___; 1151 default rel 1152 %define XMMWORD 1153 %define YMMWORD 1154 %define ZMMWORD 1155 1156 %ifdef BORINGSSL_PREFIX 1157 %include "boringssl_prefix_symbols_nasm.inc" 1158 %endif 1159 ___ 1160 } elsif ($masm) { 1161 print <<___; 1162 OPTION DOTNAME 1163 ___ 1164 } 1165 1166 if ($gas) { 1167 print <<___; 1168 #if defined(__has_feature) 1169 #if __has_feature(memory_sanitizer) && !defined(OPENSSL_NO_ASM) 1170 #define OPENSSL_NO_ASM 1171 #endif 1172 #endif 1173 1174 #if defined(__x86_64__) && !defined(OPENSSL_NO_ASM) 1175 #if defined(BORINGSSL_PREFIX) 1176 #include <boringssl_prefix_symbols_asm.h> 1177 #endif 1178 ___ 1179 } 1180 1181 while(defined(my $line=<>)) { 1182 1183 $line =~ s|\R$||; # Better chomp 1184 1185 if ($nasm) { 1186 $line =~ s|^#ifdef |%ifdef |; 1187 $line =~ s|^#ifndef |%ifndef |; 1188 $line =~ s|^#endif|%endif|; 1189 $line =~ s|[#!].*$||; # get rid of asm-style comments... 1190 } else { 1191 # Get rid of asm-style comments but not preprocessor directives. The 1192 # former are identified by having a letter after the '#' and starting in 1193 # the first column. 1194 $line =~ s|!.*$||; 1195 $line =~ s|(?<=.)#.*$||; 1196 $line =~ s|^#([^a-z].*)?$||; 1197 } 1198 1199 $line =~ s|/\*.*\*/||; # ... and C-style comments... 1200 $line =~ s|^\s+||; # ... and skip white spaces in beginning 1201 $line =~ s|\s+$||; # ... and at the end 1202 1203 if (my $label=label->re(\$line)) { print $label->out(); } 1204 1205 if (my $directive=directive->re(\$line)) { 1206 printf "%s",$directive->out(); 1207 } elsif (my $opcode=opcode->re(\$line)) { 1208 my $asm = eval("\$".$opcode->mnemonic()); 1209 1210 if ((ref($asm) eq 'CODE') && scalar(my @bytes=&$asm($line))) { 1211 print $gas?".byte\t":"DB\t",join(',',@bytes),"\n"; 1212 next; 1213 } 1214 1215 my @args; 1216 ARGUMENT: while (1) { 1217 my $arg; 1218 1219 ($arg=register->re(\$line, $opcode))|| 1220 ($arg=const->re(\$line)) || 1221 ($arg=ea->re(\$line, $opcode)) || 1222 ($arg=expr->re(\$line, $opcode)) || 1223 last ARGUMENT; 1224 1225 push @args,$arg; 1226 1227 last ARGUMENT if ($line !~ /^,/); 1228 1229 $line =~ s/^,\s*//; 1230 } # ARGUMENT: 1231 1232 if ($#args>=0) { 1233 my $insn; 1234 my $sz=$opcode->size(); 1235 1236 if ($gas) { 1237 $insn = $opcode->out($#args>=1?$args[$#args]->size():$sz); 1238 @args = map($_->out($sz),@args); 1239 printf "\t%s\t%s",$insn,join(",",@args); 1240 } else { 1241 $insn = $opcode->out(); 1242 foreach (@args) { 1243 my $arg = $_->out(); 1244 # $insn.=$sz compensates for movq, pinsrw, ... 1245 if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; } 1246 if ($arg =~ /^ymm[0-9]+$/) { $insn.=$sz; $sz="y" if(!$sz); last; } 1247 if ($arg =~ /^zmm[0-9]+$/) { $insn.=$sz; $sz="z" if(!$sz); last; } 1248 if ($arg =~ /^mm[0-9]+$/) { $insn.=$sz; $sz="q" if(!$sz); last; } 1249 } 1250 @args = reverse(@args); 1251 undef $sz if ($nasm && $opcode->mnemonic() eq "lea"); 1252 printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args)); 1253 } 1254 } else { 1255 printf "\t%s",$opcode->out(); 1256 } 1257 } 1258 1259 print $line,"\n"; 1260 } 1261 1262 print "\n$current_segment\tENDS\n" if ($current_segment && $masm); 1263 print "END\n" if ($masm); 1264 print "#endif\n" if ($gas); 1265 1266 1267 close STDOUT; 1268 1269 ################################################# 1271 # Cross-reference x86_64 ABI "card" 1272 # 1273 # Unix Win64 1274 # %rax * * 1275 # %rbx - - 1276 # %rcx #4 #1 1277 # %rdx #3 #2 1278 # %rsi #2 - 1279 # %rdi #1 - 1280 # %rbp - - 1281 # %rsp - - 1282 # %r8 #5 #3 1283 # %r9 #6 #4 1284 # %r10 * * 1285 # %r11 * * 1286 # %r12 - - 1287 # %r13 - - 1288 # %r14 - - 1289 # %r15 - - 1290 # 1291 # (*) volatile register 1292 # (-) preserved by callee 1293 # (#) Nth argument, volatile 1294 # 1295 # In Unix terms top of stack is argument transfer area for arguments 1296 # which could not be accommodated in registers. Or in other words 7th 1297 # [integer] argument resides at 8(%rsp) upon function entry point. 1298 # 128 bytes above %rsp constitute a "red zone" which is not touched 1299 # by signal handlers and can be used as temporal storage without 1300 # allocating a frame. 1301 # 1302 # In Win64 terms N*8 bytes on top of stack is argument transfer area, 1303 # which belongs to/can be overwritten by callee. N is the number of 1304 # arguments passed to callee, *but* not less than 4! This means that 1305 # upon function entry point 5th argument resides at 40(%rsp), as well 1306 # as that 32 bytes from 8(%rsp) can always be used as temporal 1307 # storage [without allocating a frame]. One can actually argue that 1308 # one can assume a "red zone" above stack pointer under Win64 as well. 1309 # Point is that at apparently no occasion Windows kernel would alter 1310 # the area above user stack pointer in true asynchronous manner... 1311 # 1312 # All the above means that if assembler programmer adheres to Unix 1313 # register and stack layout, but disregards the "red zone" existence, 1314 # it's possible to use following prologue and epilogue to "gear" from 1315 # Unix to Win64 ABI in leaf functions with not more than 6 arguments. 1316 # 1317 # omnipotent_function: 1318 # ifdef WIN64 1319 # movq %rdi,8(%rsp) 1320 # movq %rsi,16(%rsp) 1321 # movq %rcx,%rdi ; if 1st argument is actually present 1322 # movq %rdx,%rsi ; if 2nd argument is actually ... 1323 # movq %r8,%rdx ; if 3rd argument is ... 1324 # movq %r9,%rcx ; if 4th argument ... 1325 # movq 40(%rsp),%r8 ; if 5th ... 1326 # movq 48(%rsp),%r9 ; if 6th ... 1327 # endif 1328 # ... 1329 # ifdef WIN64 1330 # movq 8(%rsp),%rdi 1331 # movq 16(%rsp),%rsi 1332 # endif 1333 # ret 1334 # 1335 ################################################# 1337 # Win64 SEH, Structured Exception Handling. 1338 # 1339 # Unlike on Unix systems(*) lack of Win64 stack unwinding information 1340 # has undesired side-effect at run-time: if an exception is raised in 1341 # assembler subroutine such as those in question (basically we're 1342 # referring to segmentation violations caused by malformed input 1343 # parameters), the application is briskly terminated without invoking 1344 # any exception handlers, most notably without generating memory dump 1345 # or any user notification whatsoever. This poses a problem. It's 1346 # possible to address it by registering custom language-specific 1347 # handler that would restore processor context to the state at 1348 # subroutine entry point and return "exception is not handled, keep 1349 # unwinding" code. Writing such handler can be a challenge... But it's 1350 # doable, though requires certain coding convention. Consider following 1351 # snippet: 1352 # 1353 # .type function,@function 1354 # function: 1355 # movq %rsp,%rax # copy rsp to volatile register 1356 # pushq %r15 # save non-volatile registers 1357 # pushq %rbx 1358 # pushq %rbp 1359 # movq %rsp,%r11 1360 # subq %rdi,%r11 # prepare [variable] stack frame 1361 # andq $-64,%r11 1362 # movq %rax,0(%r11) # check for exceptions 1363 # movq %r11,%rsp # allocate [variable] stack frame 1364 # movq %rax,0(%rsp) # save original rsp value 1365 # magic_point: 1366 # ... 1367 # movq 0(%rsp),%rcx # pull original rsp value 1368 # movq -24(%rcx),%rbp # restore non-volatile registers 1369 # movq -16(%rcx),%rbx 1370 # movq -8(%rcx),%r15 1371 # movq %rcx,%rsp # restore original rsp 1372 # magic_epilogue: 1373 # ret 1374 # .size function,.-function 1375 # 1376 # The key is that up to magic_point copy of original rsp value remains 1377 # in chosen volatile register and no non-volatile register, except for 1378 # rsp, is modified. While past magic_point rsp remains constant till 1379 # the very end of the function. In this case custom language-specific 1380 # exception handler would look like this: 1381 # 1382 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, 1383 # CONTEXT *context,DISPATCHER_CONTEXT *disp) 1384 # { ULONG64 *rsp = (ULONG64 *)context->Rax; 1385 # ULONG64 rip = context->Rip; 1386 # 1387 # if (rip >= magic_point) 1388 # { rsp = (ULONG64 *)context->Rsp; 1389 # if (rip < magic_epilogue) 1390 # { rsp = (ULONG64 *)rsp[0]; 1391 # context->Rbp = rsp[-3]; 1392 # context->Rbx = rsp[-2]; 1393 # context->R15 = rsp[-1]; 1394 # } 1395 # } 1396 # context->Rsp = (ULONG64)rsp; 1397 # context->Rdi = rsp[1]; 1398 # context->Rsi = rsp[2]; 1399 # 1400 # memcpy (disp->ContextRecord,context,sizeof(CONTEXT)); 1401 # RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase, 1402 # dips->ControlPc,disp->FunctionEntry,disp->ContextRecord, 1403 # &disp->HandlerData,&disp->EstablisherFrame,NULL); 1404 # return ExceptionContinueSearch; 1405 # } 1406 # 1407 # It's appropriate to implement this handler in assembler, directly in 1408 # function's module. In order to do that one has to know members' 1409 # offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant 1410 # values. Here they are: 1411 # 1412 # CONTEXT.Rax 120 1413 # CONTEXT.Rcx 128 1414 # CONTEXT.Rdx 136 1415 # CONTEXT.Rbx 144 1416 # CONTEXT.Rsp 152 1417 # CONTEXT.Rbp 160 1418 # CONTEXT.Rsi 168 1419 # CONTEXT.Rdi 176 1420 # CONTEXT.R8 184 1421 # CONTEXT.R9 192 1422 # CONTEXT.R10 200 1423 # CONTEXT.R11 208 1424 # CONTEXT.R12 216 1425 # CONTEXT.R13 224 1426 # CONTEXT.R14 232 1427 # CONTEXT.R15 240 1428 # CONTEXT.Rip 248 1429 # CONTEXT.Xmm6 512 1430 # sizeof(CONTEXT) 1232 1431 # DISPATCHER_CONTEXT.ControlPc 0 1432 # DISPATCHER_CONTEXT.ImageBase 8 1433 # DISPATCHER_CONTEXT.FunctionEntry 16 1434 # DISPATCHER_CONTEXT.EstablisherFrame 24 1435 # DISPATCHER_CONTEXT.TargetIp 32 1436 # DISPATCHER_CONTEXT.ContextRecord 40 1437 # DISPATCHER_CONTEXT.LanguageHandler 48 1438 # DISPATCHER_CONTEXT.HandlerData 56 1439 # UNW_FLAG_NHANDLER 0 1440 # ExceptionContinueSearch 1 1441 # 1442 # In order to tie the handler to the function one has to compose 1443 # couple of structures: one for .xdata segment and one for .pdata. 1444 # 1445 # UNWIND_INFO structure for .xdata segment would be 1446 # 1447 # function_unwind_info: 1448 # .byte 9,0,0,0 1449 # .rva handler 1450 # 1451 # This structure designates exception handler for a function with 1452 # zero-length prologue, no stack frame or frame register. 1453 # 1454 # To facilitate composing of .pdata structures, auto-generated "gear" 1455 # prologue copies rsp value to rax and denotes next instruction with 1456 # .LSEH_begin_{function_name} label. This essentially defines the SEH 1457 # styling rule mentioned in the beginning. Position of this label is 1458 # chosen in such manner that possible exceptions raised in the "gear" 1459 # prologue would be accounted to caller and unwound from latter's frame. 1460 # End of function is marked with respective .LSEH_end_{function_name} 1461 # label. To summarize, .pdata segment would contain 1462 # 1463 # .rva .LSEH_begin_function 1464 # .rva .LSEH_end_function 1465 # .rva function_unwind_info 1466 # 1467 # Reference to function_unwind_info from .xdata segment is the anchor. 1468 # In case you wonder why references are 32-bit .rvas and not 64-bit 1469 # .quads. References put into these two segments are required to be 1470 # *relative* to the base address of the current binary module, a.k.a. 1471 # image base. No Win64 module, be it .exe or .dll, can be larger than 1472 # 2GB and thus such relative references can be and are accommodated in 1473 # 32 bits. 1474 # 1475 # Having reviewed the example function code, one can argue that "movq 1476 # %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix 1477 # rax would contain an undefined value. If this "offends" you, use 1478 # another register and refrain from modifying rax till magic_point is 1479 # reached, i.e. as if it was a non-volatile register. If more registers 1480 # are required prior [variable] frame setup is completed, note that 1481 # nobody says that you can have only one "magic point." You can 1482 # "liberate" non-volatile registers by denoting last stack off-load 1483 # instruction and reflecting it in finer grade unwind logic in handler. 1484 # After all, isn't it why it's called *language-specific* handler... 1485 # 1486 # SE handlers are also involved in unwinding stack when executable is 1487 # profiled or debugged. Profiling implies additional limitations that 1488 # are too subtle to discuss here. For now it's sufficient to say that 1489 # in order to simplify handlers one should either a) offload original 1490 # %rsp to stack (like discussed above); or b) if you have a register to 1491 # spare for frame pointer, choose volatile one. 1492 # 1493 # (*) Note that we're talking about run-time, not debug-time. Lack of 1494 # unwind information makes debugging hard on both Windows and 1495 # Unix. "Unlike" refers to the fact that on Unix signal handler 1496 # will always be invoked, core dumped and appropriate exit code 1497 # returned to parent (for user notification). 1498