1#! /usr/bin/env perl 2# Copyright 2005-2025 The OpenSSL Project Authors. All Rights Reserved. 3# 4# Licensed under the Apache License 2.0 (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 63 64use strict; 65 66my $flavour = shift; 67my $output = shift; 68if ($flavour =~ /\./) { $output = $flavour; undef $flavour; } 69 70open STDOUT,">$output" || die "can't open $output: $!" 71 if (defined($output)); 72 73my $gas=1; $gas=0 if ($output =~ /\.asm$/); 74my $elf=1; $elf=0 if (!$gas); 75my $win64=0; 76my $prefix=""; 77my $decor=".L"; 78 79my $masmref=8 + 50727*2**-32; # 8.00.50727 shipped with VS2005 80my $masm=0; 81my $PTR=" PTR"; 82 83my $nasmref=2.03; 84my $nasm=0; 85 86# GNU as indicator, as opposed to $gas, which indicates acceptable 87# syntax 88my $gnuas=0; 89 90if ($flavour eq "mingw64") { $gas=1; $elf=0; $win64=1; 91 $prefix=`echo __USER_LABEL_PREFIX__ | $ENV{CC} -E -P -`; 92 $prefix =~ s|\R$||; # Better chomp 93 } 94elsif ($flavour eq "macosx") { $gas=1; $elf=0; $prefix="_"; $decor="L\$"; } 95elsif ($flavour eq "masm") { $gas=0; $elf=0; $masm=$masmref; $win64=1; $decor="\$L\$"; } 96elsif ($flavour eq "nasm") { $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $decor="\$L\$"; $PTR=""; } 97elsif (!$gas) 98{ if ($ENV{ASM} =~ m/nasm/ && `nasm -v` =~ m/version ([0-9]+)\.([0-9]+)/i) 99 { $nasm = $1 + $2*0.01; $PTR=""; } 100 elsif (`ml64 2>&1` =~ m/Version ([0-9]+)\.([0-9]+)(\.([0-9]+))?/) 101 { $masm = $1 + $2*2**-16 + $4*2**-32; } 102 die "no assembler found on %PATH%" if (!($nasm || $masm)); 103 $win64=1; 104 $elf=0; 105 $decor="\$L\$"; 106} 107# Find out if we're using GNU as 108elsif (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1` 109 =~ /GNU assembler version ([2-9]\.[0-9]+)/) 110{ 111 $gnuas=1; 112} 113elsif (`$ENV{CC} --version 2>/dev/null` 114 =~ /(clang .*|Intel.*oneAPI .*)/) 115{ 116 $gnuas=1; 117} 118elsif (`$ENV{CC} -V 2>/dev/null` 119 =~ /nvc .*/) 120{ 121 $gnuas=1; 122} 123 124my $cet_property; 125if ($flavour =~ /elf/) { 126 # Always generate .note.gnu.property section for ELF outputs to 127 # mark Intel CET support since all input files must be marked 128 # with Intel CET support in order for linker to mark output with 129 # Intel CET support. 130 my $p2align=3; $p2align=2 if ($flavour eq "elf32"); 131 my $section='.note.gnu.property, #alloc'; 132 $section='".note.gnu.property", "a"' if $gnuas; 133 $cet_property = <<_____; 134 .section $section 135 .p2align $p2align 136 .long 1f - 0f 137 .long 4f - 1f 138 .long 5 1390: 140 # "GNU" encoded with .byte, since .asciz isn't supported 141 # on Solaris. 142 .byte 0x47 143 .byte 0x4e 144 .byte 0x55 145 .byte 0 1461: 147 .p2align $p2align 148 .long 0xc0000002 149 .long 3f - 2f 1502: 151 .long 3 1523: 153 .p2align $p2align 1544: 155_____ 156} 157 158my $current_segment; 159# 160# I could not find equivalent of .previous directive for MASM (Microsoft 161# assembler ML). Using of .previous got introduced to .pl files with 162# placing of various constants into .rodata sections (segments). 163# Each .rodata section is terminated by .previous directive which 164# restores the preceding section to .rodata: 165# 166# .text 167# ; this is is the text section/segment 168# .rodata 169# ; constant definitions go here 170# .previous 171# ; the .text section which precedes .rodata got restored here 172# 173# The equivalent form for masm reads as follows: 174# 175# .text$ SEGMENT ALIGN(256) 'CODE' 176# ; this is is the text section/segment 177# .text$ ENDS 178# .rdata SEGMENT READONLY ALIGN(64) 179# ; constant definitions go here 180# .rdata$ ENDS 181# .text$ SEGMENT ALIGN(256) 'CODE' 182# ; text section follows 183# .text$ ENDS 184# 185# The .previous directive typically terminates .roadata segments/sections which 186# hold definitions of constants. In order to place constants into .rdata 187# segments when using masm we need to introduce a segment_stack array so we can 188# emit proper ENDS directive whenever we see .previous. 189# 190# The code is tailored to work current set of .pl/asm files. There are some 191# inconsistencies. For example .text section is the first section in all those 192# files except ecp_nistz256. So we need to take that into account. 193# 194# ; stack is empty 195# .text 196# ; push '.text ' section twice, the stack looks as 197# ; follows: 198# ; ('.text', '.text') 199# .rodata 200# ; pop() so we can generate proper 'ENDS' for masm. 201# ; stack looks like: 202# ; ('.text') 203# ; push '.rodata', so we can create corresponding ENDS for masm. 204# ; stack looks like: 205# ; ('.rodata', '.text') 206# .previous 207# ; pop() '.rodata' from stack, so we create '.rodata ENDS' 208# ; in masm flavour. For nasm flavour we just pop() because 209# ; nasm does not use .rodata ENDS to close the current section 210# ; the stack content is like this: 211# ; ('.text', '.text') 212# ; pop() again to find a previous section we need to restore. 213# ; Depending on flavour we either generate .section .text 214# ; or .text SEGMENT. The stack looks like: 215# ; ('.text') 216# 217my @segment_stack = (); 218my $current_function; 219my %globals; 220 221{ package vex_prefix; # pick up vex prefixes, example: {vex} vpmadd52luq m256, %ymm, %ymm 222 sub re { 223 my ($class, $line) = @_; 224 my $self = {}; 225 my $ret; 226 227 if ($$line =~ /(^\{vex\})/) { 228 bless $self,$class; 229 $self->{value} = $1; 230 $ret = $self; 231 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 232 } 233 $ret; 234 } 235 sub out { 236 my $self = shift; 237 $self->{value}; 238 } 239} 240{ package opcode; # pick up opcodes 241 sub re { 242 my ($class, $line) = @_; 243 my $self = {}; 244 my $ret; 245 246 if ($$line =~ /^([a-z][a-z0-9]*)/i) { 247 bless $self,$class; 248 $self->{op} = $1; 249 $ret = $self; 250 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 251 252 undef $self->{sz}; 253 if ($self->{op} =~ /^(movz)x?([bw]).*/) { # movz is pain... 254 $self->{op} = $1; 255 $self->{sz} = $2; 256 } elsif ($self->{op} =~ /call|jmp/) { 257 $self->{sz} = ""; 258 } elsif ($self->{op} =~ /^p/ && $' !~ /^(ush|op|insrw)/) { # SSEn 259 $self->{sz} = ""; 260 } elsif ($self->{op} =~ /^[vk]/) { # VEX or k* such as kmov 261 $self->{sz} = ""; 262 } elsif ($self->{op} =~ /mov[dq]/ && $$line =~ /%xmm/) { 263 $self->{sz} = ""; 264 } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) { 265 $self->{op} = $1; 266 $self->{sz} = $2; 267 } 268 } 269 $ret; 270 } 271 sub size { 272 my ($self, $sz) = @_; 273 $self->{sz} = $sz if (defined($sz) && !defined($self->{sz})); 274 $self->{sz}; 275 } 276 sub out { 277 my $self = shift; 278 if ($gas) { 279 if ($self->{op} eq "movz") { # movz is pain... 280 sprintf "%s%s%s",$self->{op},$self->{sz},shift; 281 } elsif ($self->{op} =~ /^set/) { 282 "$self->{op}"; 283 } elsif ($self->{op} eq "ret") { 284 my $epilogue = ""; 285 if ($win64 && $current_function->{abi} eq "svr4") { 286 $epilogue = "movq 8(%rsp),%rdi\n\t" . 287 "movq 16(%rsp),%rsi\n\t"; 288 } 289 $epilogue . ".byte 0xf3,0xc3"; 290 } elsif ($self->{op} eq "call" && !$elf && $current_segment eq ".init") { 291 ".p2align\t3\n\t.quad"; 292 } else { 293 "$self->{op}$self->{sz}"; 294 } 295 } else { 296 $self->{op} =~ s/^movz/movzx/; 297 if ($self->{op} eq "ret") { 298 $self->{op} = ""; 299 if ($win64 && $current_function->{abi} eq "svr4") { 300 $self->{op} = "mov rdi,QWORD$PTR\[8+rsp\]\t;WIN64 epilogue\n\t". 301 "mov rsi,QWORD$PTR\[16+rsp\]\n\t"; 302 } 303 $self->{op} .= "DB\t0F3h,0C3h\t\t;repret"; 304 } elsif ($self->{op} =~ /^(pop|push)f/) { 305 $self->{op} .= $self->{sz}; 306 } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") { 307 $self->{op} = "\tDQ"; 308 } 309 $self->{op}; 310 } 311 } 312 sub mnemonic { 313 my ($self, $op) = @_; 314 $self->{op}=$op if (defined($op)); 315 $self->{op}; 316 } 317} 318{ package const; # pick up constants, which start with $ 319 sub re { 320 my ($class, $line) = @_; 321 my $self = {}; 322 my $ret; 323 324 if ($$line =~ /^\$([^,]+)/) { 325 bless $self, $class; 326 $self->{value} = $1; 327 $ret = $self; 328 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 329 } 330 $ret; 331 } 332 sub out { 333 my $self = shift; 334 335 $self->{value} =~ s/\b(0b[0-1]+)/oct($1)/eig; 336 if ($gas) { 337 # Solaris /usr/ccs/bin/as can't handle multiplications 338 # in $self->{value} 339 my $value = $self->{value}; 340 no warnings; # oct might complain about overflow, ignore here... 341 $value =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi; 342 if ($value =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg) { 343 $self->{value} = $value; 344 } 345 sprintf "\$%s",$self->{value}; 346 } else { 347 my $value = $self->{value}; 348 $value =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm); 349 sprintf "%s",$value; 350 } 351 } 352} 353{ package ea; # pick up effective addresses: expr(%reg,%reg,scale) 354 355 my %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR", 356 l=>"DWORD$PTR", d=>"DWORD$PTR", 357 q=>"QWORD$PTR", o=>"OWORD$PTR", 358 x=>"XMMWORD$PTR", y=>"YMMWORD$PTR", 359 z=>"ZMMWORD$PTR" ) if (!$gas); 360 361 sub re { 362 my ($class, $line, $opcode) = @_; 363 my $self = {}; 364 my $ret; 365 366 # optional * ----vvv--- appears in indirect jmp/call 367 if ($$line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)((?:{[^}]+})*)/) { 368 bless $self, $class; 369 $self->{asterisk} = $1; 370 $self->{label} = $2; 371 ($self->{base},$self->{index},$self->{scale})=split(/,/,$3); 372 $self->{scale} = 1 if (!defined($self->{scale})); 373 $self->{opmask} = $4; 374 $ret = $self; 375 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 376 377 if ($win64 && $self->{label} =~ s/\@GOTPCREL//) { 378 die if ($opcode->mnemonic() ne "mov"); 379 $opcode->mnemonic("lea"); 380 } 381 $self->{base} =~ s/^%//; 382 $self->{index} =~ s/^%// if (defined($self->{index})); 383 $self->{opcode} = $opcode; 384 } 385 $ret; 386 } 387 sub size {} 388 sub out { 389 my ($self, $sz) = @_; 390 391 $self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 392 $self->{label} =~ s/\.L/$decor/g; 393 394 # Silently convert all EAs to 64-bit. This is required for 395 # elder GNU assembler and results in more compact code, 396 # *but* most importantly AES module depends on this feature! 397 $self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; 398 $self->{base} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; 399 400 # Solaris /usr/ccs/bin/as can't handle multiplications 401 # in $self->{label}... 402 use integer; 403 $self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi; 404 $self->{label} =~ s/\b([0-9]+\s*[\*\/\%]\s*[0-9]+)\b/eval($1)/eg; 405 406 # Some assemblers insist on signed presentation of 32-bit 407 # offsets, but sign extension is a tricky business in perl... 408 if ((1<<31)<<1) { 409 $self->{label} =~ s/\b([0-9]+)\b/$1<<32>>32/eg; 410 } else { 411 $self->{label} =~ s/\b([0-9]+)\b/$1>>0/eg; 412 } 413 414 # if base register is %rbp or %r13, see if it's possible to 415 # flip base and index registers [for better performance] 416 if (!$self->{label} && $self->{index} && $self->{scale}==1 && 417 $self->{base} =~ /(rbp|r13)/) { 418 $self->{base} = $self->{index}; $self->{index} = $1; 419 } 420 421 if ($gas) { 422 $self->{label} =~ s/^___imp_/__imp__/ if ($flavour eq "mingw64"); 423 424 if (defined($self->{index})) { 425 sprintf "%s%s(%s,%%%s,%d)%s", 426 $self->{asterisk},$self->{label}, 427 $self->{base}?"%$self->{base}":"", 428 $self->{index},$self->{scale}, 429 $self->{opmask}; 430 } else { 431 sprintf "%s%s(%%%s)%s", $self->{asterisk},$self->{label}, 432 $self->{base},$self->{opmask}; 433 } 434 } else { 435 $self->{label} =~ s/\./\$/g; 436 $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig; 437 $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/); 438 439 my $mnemonic = $self->{opcode}->mnemonic(); 440 ($self->{asterisk}) && ($sz="q") || 441 ($mnemonic =~ /^v?mov([qd])$/) && ($sz=$1) || 442 ($mnemonic =~ /^v?pinsr([qdwb])$/) && ($sz=$1) || 443 ($mnemonic =~ /^vpbroadcast([qdwb])$/) && ($sz=$1) || 444 ($mnemonic =~ /^v(?!perm)[a-z]+[fi]128$/) && ($sz="x"); 445 446 $self->{opmask} =~ s/%(k[0-7])/$1/; 447 448 if (defined($self->{index})) { 449 sprintf "%s[%s%s*%d%s]%s",$szmap{$sz}, 450 $self->{label}?"$self->{label}+":"", 451 $self->{index},$self->{scale}, 452 $self->{base}?"+$self->{base}":"", 453 $self->{opmask}; 454 } elsif ($self->{base} eq "rip") { 455 sprintf "%s[%s]",$szmap{$sz},$self->{label}; 456 } else { 457 sprintf "%s[%s%s]%s", $szmap{$sz}, 458 $self->{label}?"$self->{label}+":"", 459 $self->{base},$self->{opmask}; 460 } 461 } 462 } 463} 464{ package register; # pick up registers, which start with %. 465 sub re { 466 my ($class, $line, $opcode) = @_; 467 my $self = {}; 468 my $ret; 469 470 # optional * ----vvv--- appears in indirect jmp/call 471 if ($$line =~ /^(\*?)%(\w+)((?:{[^}]+})*)/) { 472 bless $self,$class; 473 $self->{asterisk} = $1; 474 $self->{value} = $2; 475 $self->{opmask} = $3; 476 $opcode->size($self->size()); 477 $ret = $self; 478 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 479 } 480 $ret; 481 } 482 sub size { 483 my $self = shift; 484 my $ret; 485 486 if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; } 487 elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; } 488 elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; } 489 elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; } 490 elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; } 491 elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; } 492 elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; } 493 elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; } 494 495 $ret; 496 } 497 sub out { 498 my $self = shift; 499 if ($gas) { sprintf "%s%%%s%s", $self->{asterisk}, 500 $self->{value}, 501 $self->{opmask}; } 502 else { $self->{opmask} =~ s/%(k[0-7])/$1/; 503 $self->{value}.$self->{opmask}; } 504 } 505} 506{ package label; # pick up labels, which end with : 507 sub re { 508 my ($class, $line) = @_; 509 my $self = {}; 510 my $ret; 511 512 if ($$line =~ /(^[\.\w]+)\:/) { 513 bless $self,$class; 514 $self->{value} = $1; 515 $ret = $self; 516 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 517 518 $self->{value} =~ s/^\.L/$decor/; 519 } 520 $ret; 521 } 522 sub out { 523 my $self = shift; 524 525 if ($gas) { 526 my $func = ($globals{$self->{value}} or $self->{value}) . ":"; 527 if ($win64 && $current_function->{name} eq $self->{value} 528 && $current_function->{abi} eq "svr4") { 529 $func .= "\n"; 530 $func .= " movq %rdi,8(%rsp)\n"; 531 $func .= " movq %rsi,16(%rsp)\n"; 532 $func .= " movq %rsp,%rax\n"; 533 $func .= "${decor}SEH_begin_$current_function->{name}:\n"; 534 my $narg = $current_function->{narg}; 535 $narg=6 if (!defined($narg)); 536 $func .= " movq %rcx,%rdi\n" if ($narg>0); 537 $func .= " movq %rdx,%rsi\n" if ($narg>1); 538 $func .= " movq %r8,%rdx\n" if ($narg>2); 539 $func .= " movq %r9,%rcx\n" if ($narg>3); 540 $func .= " movq 40(%rsp),%r8\n" if ($narg>4); 541 $func .= " movq 48(%rsp),%r9\n" if ($narg>5); 542 } 543 $func; 544 } elsif ($self->{value} ne "$current_function->{name}") { 545 # Make all labels in masm global. 546 $self->{value} .= ":" if ($masm); 547 $self->{value} . ":"; 548 } elsif ($win64 && $current_function->{abi} eq "svr4") { 549 my $func = "$current_function->{name}" . 550 ($nasm ? ":" : "\tPROC $current_function->{scope}") . 551 "\n"; 552 $func .= " mov QWORD$PTR\[8+rsp\],rdi\t;WIN64 prologue\n"; 553 $func .= " mov QWORD$PTR\[16+rsp\],rsi\n"; 554 $func .= " mov rax,rsp\n"; 555 $func .= "${decor}SEH_begin_$current_function->{name}:"; 556 $func .= ":" if ($masm); 557 $func .= "\n"; 558 my $narg = $current_function->{narg}; 559 $narg=6 if (!defined($narg)); 560 $func .= " mov rdi,rcx\n" if ($narg>0); 561 $func .= " mov rsi,rdx\n" if ($narg>1); 562 $func .= " mov rdx,r8\n" if ($narg>2); 563 $func .= " mov rcx,r9\n" if ($narg>3); 564 $func .= " mov r8,QWORD$PTR\[40+rsp\]\n" if ($narg>4); 565 $func .= " mov r9,QWORD$PTR\[48+rsp\]\n" if ($narg>5); 566 $func .= "\n"; 567 } else { 568 "$current_function->{name}". 569 ($nasm ? ":" : "\tPROC $current_function->{scope}"); 570 } 571 } 572} 573{ package expr; # pick up expressions 574 sub re { 575 my ($class, $line, $opcode) = @_; 576 my $self = {}; 577 my $ret; 578 579 if ($$line =~ /(^[^,]+)/) { 580 bless $self,$class; 581 $self->{value} = $1; 582 $ret = $self; 583 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 584 585 $self->{value} =~ s/\@PLT// if (!$elf); 586 $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 587 $self->{value} =~ s/\.L/$decor/g; 588 $self->{opcode} = $opcode; 589 } 590 $ret; 591 } 592 sub out { 593 my $self = shift; 594 if ($nasm && $self->{opcode}->mnemonic()=~m/^j(?![re]cxz)/) { 595 "NEAR ".$self->{value}; 596 } else { 597 $self->{value}; 598 } 599 } 600} 601{ package cfi_directive; 602 # CFI directives annotate instructions that are significant for 603 # stack unwinding procedure compliant with DWARF specification, 604 # see http://dwarfstd.org/. Besides naturally expected for this 605 # script platform-specific filtering function, this module adds 606 # three auxiliary synthetic directives not recognized by [GNU] 607 # assembler: 608 # 609 # - .cfi_push to annotate push instructions in prologue, which 610 # translates to .cfi_adjust_cfa_offset (if needed) and 611 # .cfi_offset; 612 # - .cfi_pop to annotate pop instructions in epilogue, which 613 # translates to .cfi_adjust_cfa_offset (if needed) and 614 # .cfi_restore; 615 # - [and most notably] .cfi_cfa_expression which encodes 616 # DW_CFA_def_cfa_expression and passes it to .cfi_escape as 617 # byte vector; 618 # 619 # CFA expressions were introduced in DWARF specification version 620 # 3 and describe how to deduce CFA, Canonical Frame Address. This 621 # becomes handy if your stack frame is variable and you can't 622 # spare register for [previous] frame pointer. Suggested directive 623 # syntax is made-up mix of DWARF operator suffixes [subset of] 624 # and references to registers with optional bias. Following example 625 # describes offloaded *original* stack pointer at specific offset 626 # from *current* stack pointer: 627 # 628 # .cfi_cfa_expression %rsp+40,deref,+8 629 # 630 # Final +8 has everything to do with the fact that CFA is defined 631 # as reference to top of caller's stack, and on x86_64 call to 632 # subroutine pushes 8-byte return address. In other words original 633 # stack pointer upon entry to a subroutine is 8 bytes off from CFA. 634 635 # Below constants are taken from "DWARF Expressions" section of the 636 # DWARF specification, section is numbered 7.7 in versions 3 and 4. 637 my %DW_OP_simple = ( # no-arg operators, mapped directly 638 deref => 0x06, dup => 0x12, 639 drop => 0x13, over => 0x14, 640 pick => 0x15, swap => 0x16, 641 rot => 0x17, xderef => 0x18, 642 643 abs => 0x19, and => 0x1a, 644 div => 0x1b, minus => 0x1c, 645 mod => 0x1d, mul => 0x1e, 646 neg => 0x1f, not => 0x20, 647 or => 0x21, plus => 0x22, 648 shl => 0x24, shr => 0x25, 649 shra => 0x26, xor => 0x27, 650 ); 651 652 my %DW_OP_complex = ( # used in specific subroutines 653 constu => 0x10, # uleb128 654 consts => 0x11, # sleb128 655 plus_uconst => 0x23, # uleb128 656 lit0 => 0x30, # add 0-31 to opcode 657 reg0 => 0x50, # add 0-31 to opcode 658 breg0 => 0x70, # add 0-31 to opcole, sleb128 659 regx => 0x90, # uleb28 660 fbreg => 0x91, # sleb128 661 bregx => 0x92, # uleb128, sleb128 662 piece => 0x93, # uleb128 663 ); 664 665 # Following constants are defined in x86_64 ABI supplement, for 666 # example available at https://www.uclibc.org/docs/psABI-x86_64.pdf, 667 # see section 3.7 "Stack Unwind Algorithm". 668 my %DW_reg_idx = ( 669 "%rax"=>0, "%rdx"=>1, "%rcx"=>2, "%rbx"=>3, 670 "%rsi"=>4, "%rdi"=>5, "%rbp"=>6, "%rsp"=>7, 671 "%r8" =>8, "%r9" =>9, "%r10"=>10, "%r11"=>11, 672 "%r12"=>12, "%r13"=>13, "%r14"=>14, "%r15"=>15 673 ); 674 675 my ($cfa_reg, $cfa_rsp); 676 my @cfa_stack; 677 678 # [us]leb128 format is variable-length integer representation base 679 # 2^128, with most significant bit of each byte being 0 denoting 680 # *last* most significant digit. See "Variable Length Data" in the 681 # DWARF specification, numbered 7.6 at least in versions 3 and 4. 682 sub sleb128 { 683 use integer; # get right shift extend sign 684 685 my $val = shift; 686 my $sign = ($val < 0) ? -1 : 0; 687 my @ret = (); 688 689 while(1) { 690 push @ret, $val&0x7f; 691 692 # see if remaining bits are same and equal to most 693 # significant bit of the current digit, if so, it's 694 # last digit... 695 last if (($val>>6) == $sign); 696 697 @ret[-1] |= 0x80; 698 $val >>= 7; 699 } 700 701 return @ret; 702 } 703 sub uleb128 { 704 my $val = shift; 705 my @ret = (); 706 707 while(1) { 708 push @ret, $val&0x7f; 709 710 # see if it's last significant digit... 711 last if (($val >>= 7) == 0); 712 713 @ret[-1] |= 0x80; 714 } 715 716 return @ret; 717 } 718 sub const { 719 my $val = shift; 720 721 if ($val >= 0 && $val < 32) { 722 return ($DW_OP_complex{lit0}+$val); 723 } 724 return ($DW_OP_complex{consts}, sleb128($val)); 725 } 726 sub reg { 727 my $val = shift; 728 729 return if ($val !~ m/^(%r\w+)(?:([\+\-])((?:0x)?[0-9a-f]+))?/); 730 731 my $reg = $DW_reg_idx{$1}; 732 my $off = eval ("0 $2 $3"); 733 734 return (($DW_OP_complex{breg0} + $reg), sleb128($off)); 735 # Yes, we use DW_OP_bregX+0 to push register value and not 736 # DW_OP_regX, because latter would require even DW_OP_piece, 737 # which would be a waste under the circumstances. If you have 738 # to use DWP_OP_reg, use "regx:N"... 739 } 740 sub cfa_expression { 741 my $line = shift; 742 my @ret; 743 744 foreach my $token (split(/,\s*/,$line)) { 745 if ($token =~ /^%r/) { 746 push @ret,reg($token); 747 } elsif ($token =~ /((?:0x)?[0-9a-f]+)\((%r\w+)\)/) { 748 push @ret,reg("$2+$1"); 749 } elsif ($token =~ /(\w+):(\-?(?:0x)?[0-9a-f]+)(U?)/i) { 750 my $i = 1*eval($2); 751 push @ret,$DW_OP_complex{$1}, ($3 ? uleb128($i) : sleb128($i)); 752 } elsif (my $i = 1*eval($token) or $token eq "0") { 753 if ($token =~ /^\+/) { 754 push @ret,$DW_OP_complex{plus_uconst},uleb128($i); 755 } else { 756 push @ret,const($i); 757 } 758 } else { 759 push @ret,$DW_OP_simple{$token}; 760 } 761 } 762 763 # Finally we return DW_CFA_def_cfa_expression, 15, followed by 764 # length of the expression and of course the expression itself. 765 return (15,scalar(@ret),@ret); 766 } 767 sub re { 768 my ($class, $line) = @_; 769 my $self = {}; 770 my $ret; 771 772 if ($$line =~ s/^\s*\.cfi_(\w+)\s*//) { 773 bless $self,$class; 774 $ret = $self; 775 undef $self->{value}; 776 my $dir = $1; 777 778 SWITCH: for ($dir) { 779 # What is $cfa_rsp? Effectively it's difference between %rsp 780 # value and current CFA, Canonical Frame Address, which is 781 # why it starts with -8. Recall that CFA is top of caller's 782 # stack... 783 /startproc/ && do { ($cfa_reg, $cfa_rsp) = ("%rsp", -8); last; }; 784 /endproc/ && do { ($cfa_reg, $cfa_rsp) = ("%rsp", 0); 785 # .cfi_remember_state directives that are not 786 # matched with .cfi_restore_state are 787 # unnecessary. 788 die "unpaired .cfi_remember_state" if (@cfa_stack); 789 last; 790 }; 791 /def_cfa_register/ 792 && do { $cfa_reg = $$line; last; }; 793 /def_cfa_offset/ 794 && do { $cfa_rsp = -1*eval($$line) if ($cfa_reg eq "%rsp"); 795 last; 796 }; 797 /adjust_cfa_offset/ 798 && do { $cfa_rsp -= 1*eval($$line) if ($cfa_reg eq "%rsp"); 799 last; 800 }; 801 /def_cfa/ && do { if ($$line =~ /(%r\w+)\s*,\s*(.+)/) { 802 $cfa_reg = $1; 803 $cfa_rsp = -1*eval($2) if ($cfa_reg eq "%rsp"); 804 } 805 last; 806 }; 807 /push/ && do { $dir = undef; 808 $cfa_rsp -= 8; 809 if ($cfa_reg eq "%rsp") { 810 $self->{value} = ".cfi_adjust_cfa_offset\t8\n"; 811 } 812 $self->{value} .= ".cfi_offset\t$$line,$cfa_rsp"; 813 last; 814 }; 815 /pop/ && do { $dir = undef; 816 $cfa_rsp += 8; 817 if ($cfa_reg eq "%rsp") { 818 $self->{value} = ".cfi_adjust_cfa_offset\t-8\n"; 819 } 820 $self->{value} .= ".cfi_restore\t$$line"; 821 last; 822 }; 823 /cfa_expression/ 824 && do { $dir = undef; 825 $self->{value} = ".cfi_escape\t" . 826 join(",", map(sprintf("0x%02x", $_), 827 cfa_expression($$line))); 828 last; 829 }; 830 /remember_state/ 831 && do { push @cfa_stack, [$cfa_reg, $cfa_rsp]; 832 last; 833 }; 834 /restore_state/ 835 && do { ($cfa_reg, $cfa_rsp) = @{pop @cfa_stack}; 836 last; 837 }; 838 } 839 840 $self->{value} = ".cfi_$dir\t$$line" if ($dir); 841 842 $$line = ""; 843 } 844 845 return $ret; 846 } 847 sub out { 848 my $self = shift; 849 return ($elf ? $self->{value} : undef); 850 } 851} 852{ package directive; # pick up directives, which start with . 853 sub re { 854 my ($class, $line) = @_; 855 my $self = {}; 856 my $ret; 857 my $dir; 858 859 # chain-call to cfi_directive 860 $ret = cfi_directive->re($line) and return $ret; 861 862 if ($$line =~ /^\s*(\.\w+)/) { 863 bless $self,$class; 864 $dir = $1; 865 $ret = $self; 866 undef $self->{value}; 867 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 868 869 SWITCH: for ($dir) { 870 /\.global|\.globl|\.extern/ 871 && do { $globals{$$line} = $prefix . $$line; 872 $$line = $globals{$$line} if ($prefix); 873 last; 874 }; 875 /\.type/ && do { my ($sym,$type,$narg) = split(',',$$line); 876 if ($type eq "\@function") { 877 undef $current_function; 878 $current_function->{name} = $sym; 879 $current_function->{abi} = "svr4"; 880 $current_function->{narg} = $narg; 881 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 882 } elsif ($type eq "\@abi-omnipotent") { 883 undef $current_function; 884 $current_function->{name} = $sym; 885 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 886 } 887 $$line =~ s/\@abi\-omnipotent/\@function/; 888 $$line =~ s/\@function.*/\@function/; 889 last; 890 }; 891 /\.asciz/ && do { if ($$line =~ /^"(.*)"$/) { 892 $dir = ".byte"; 893 $$line = join(",",unpack("C*",$1),0); 894 } 895 last; 896 }; 897 /\.rva|\.long|\.quad|\.byte/ 898 && do { $$line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 899 $$line =~ s/\.L/$decor/g; 900 last; 901 }; 902 } 903 904 if ($gas) { 905 $self->{value} = $dir . "\t" . $$line; 906 907 if ($dir =~ /\.extern/) { 908 $self->{value} = ""; # swallow extern 909 } elsif (!$elf && $dir =~ /\.type/) { 910 $self->{value} = ""; 911 $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" . 912 (defined($globals{$1})?".scl 2;":".scl 3;") . 913 "\t.type 32;\t.endef" 914 if ($win64 && $$line =~ /([^,]+),\@function/); 915 } elsif (!$elf && $dir =~ /\.size/) { 916 $self->{value} = ""; 917 if (defined($current_function)) { 918 $self->{value} .= "${decor}SEH_end_$current_function->{name}:" 919 if ($win64 && $current_function->{abi} eq "svr4"); 920 undef $current_function; 921 } 922 } elsif (!$elf && $dir =~ /\.align/) { 923 $self->{value} = ".p2align\t" . (log($$line)/log(2)); 924 } elsif ($dir eq ".section") { 925 # 926 # get rid off align option, it's not supported/tolerated 927 # by gcc. openssl project introduced the option as an aid 928 # to deal with nasm/masm assembly. 929 # 930 $self->{value} =~ s/(.+)\s+align\s*=.*$/$1/; 931 $current_segment = pop(@segment_stack); 932 if (not $current_segment) { 933 # if no previous section is defined, then assume .text 934 # so code does not land in .data section by accident. 935 # this deals with inconsistency of perl-assembly files. 936 push(@segment_stack, ".text"); 937 } 938 # 939 # $$line may still contains align= option. We do care 940 # about section type here. 941 # 942 $current_segment = $$line; 943 $current_segment =~ s/([^\s]+).*$/$1/; 944 push(@segment_stack, $current_segment); 945 if (!$elf && $current_segment eq ".rodata") { 946 if ($flavour eq "macosx") { $self->{value} = ".section\t__DATA,__const"; } 947 elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.rodata"; } 948 } 949 if (!$elf && $current_segment eq ".init") { 950 if ($flavour eq "macosx") { $self->{value} = ".mod_init_func"; } 951 elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.ctors"; } 952 } 953 } elsif ($dir =~ /\.(text|data)/) { 954 $current_segment = pop(@segment_stack); 955 if (not $current_segment) { 956 # if no previous section is defined, then assume .text 957 # so code does not land in .data section by accident. 958 # this deals with inconsistency of perl-assembly files. 959 push(@segment_stack, ".text"); 960 } 961 $current_segment=".$1"; 962 push(@segment_stack, $current_segment); 963 } elsif ($dir =~ /\.hidden/) { 964 if ($flavour eq "macosx") { $self->{value} = ".private_extern\t$prefix$$line"; } 965 elsif ($flavour eq "mingw64") { $self->{value} = ""; } 966 } elsif ($dir =~ /\.comm/) { 967 $self->{value} = "$dir\t$prefix$$line"; 968 $self->{value} =~ s|,([0-9]+),([0-9]+)$|",$1,".log($2)/log(2)|e if ($flavour eq "macosx"); 969 } elsif ($dir =~ /\.previous/) { 970 pop(@segment_stack); #pop ourselves 971 # just peek at the top of the stack here 972 $current_segment = @segment_stack[0]; 973 if (not $current_segment) { 974 # if no previous segment was defined assume .text so 975 # the code does not accidentally land in .data section. 976 $current_segment = ".text"; 977 push(@segment_stack, $current_segment); 978 } 979 if ($flavour eq "mingw64" || $flavour eq "macosx") { 980 $self->{value} = $current_segment; 981 } 982 } 983 $$line = ""; 984 return $self; 985 } 986 987 # non-gas case or nasm/masm 988 SWITCH: for ($dir) { 989 /\.text/ && do { my $v=undef; 990 if ($nasm) { 991 $current_segment = pop(@segment_stack); 992 if (not $current_segment) { 993 push(@segment_stack, ".text"); 994 } 995 $v="section .text code align=64\n"; 996 $current_segment = ".text"; 997 push(@segment_stack, $current_segment); 998 } else { 999 $current_segment = pop(@segment_stack); 1000 if (not $current_segment) { 1001 push(@segment_stack, ".text\$"); 1002 } 1003 $v="$current_segment\tENDS\n" if ($current_segment); 1004 $current_segment = ".text\$"; 1005 push(@segment_stack, $current_segment); 1006 $v.="$current_segment\tSEGMENT "; 1007 $v.=$masm>=$masmref ? "ALIGN(256)" : "PAGE"; 1008 $v.=" 'CODE'"; 1009 } 1010 $self->{value} = $v; 1011 last; 1012 }; 1013 /\.data/ && do { my $v=undef; 1014 if ($nasm) { 1015 $v="section .data data align=8\n"; 1016 } else { 1017 $current_segment = pop(@segment_stack); 1018 $v="$current_segment\tENDS\n" if ($current_segment); 1019 $current_segment = "_DATA"; 1020 push(@segment_stack, $current_segment); 1021 $v.="$current_segment\tSEGMENT"; 1022 } 1023 $self->{value} = $v; 1024 last; 1025 }; 1026 /\.section/ && do { my $v=undef; 1027 my $align=undef; 1028 # 1029 # $$line may currently contain something like this 1030 # .rodata align = 64 1031 # align part is optional 1032 # 1033 $align = $$line; 1034 $align =~ s/(.*)(align\s*=\s*\d+$)/$2/; 1035 $$line =~ s/(.*)(\s+align\s*=\s*\d+$)/$1/; 1036 $$line =~ s/,.*//; 1037 $$line = ".CRT\$XCU" if ($$line eq ".init"); 1038 $$line = ".rdata" if ($$line eq ".rodata"); 1039 if ($nasm) { 1040 $current_segment = pop(@segment_stack); 1041 if (not $current_segment) { 1042 # 1043 # This is a hack which deals with ecp_nistz256-x86_64.pl, 1044 # The precomputed curve is stored in the first section 1045 # in .asm file. Pushing extra .text section here 1046 # allows our poor man's solution to stick to assumption 1047 # .text section is always the first. 1048 # 1049 push(@segment_stack, ".text"); 1050 } 1051 $v="section $$line"; 1052 if ($$line=~/\.([prx])data/) { 1053 if ($align =~ /align\s*=\s*(\d+)/) { 1054 $v.= " rdata align=$1" ; 1055 } else { 1056 $v.=" rdata align="; 1057 $v.=$1 eq "p"? 4 : 8; 1058 } 1059 } elsif ($$line=~/\.CRT\$/i) { 1060 $v.=" rdata align=8"; 1061 } 1062 } else { 1063 $current_segment = pop(@segment_stack); 1064 if (not $current_segment) { 1065 # 1066 # same hack for masm to keep ecp_nistz256-x86_64.pl 1067 # happy. 1068 # 1069 push(@segment_stack, ".text\$"); 1070 } 1071 $v="$current_segment\tENDS\n" if ($current_segment); 1072 $v.="$$line\tSEGMENT"; 1073 if ($$line=~/\.([prx])data/) { 1074 $v.=" READONLY"; 1075 if ($align =~ /align\s*=\s*(\d+)$/) { 1076 $v.=" ALIGN($1)" if ($masm>=$masmref); 1077 } else { 1078 $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref); 1079 } 1080 } elsif ($$line=~/\.CRT\$/i) { 1081 $v.=" READONLY "; 1082 $v.=$masm>=$masmref ? "ALIGN(8)" : "DWORD"; 1083 } 1084 } 1085 $current_segment = $$line; 1086 push(@segment_stack, $$line); 1087 $self->{value} = $v; 1088 last; 1089 }; 1090 /\.extern/ && do { $self->{value} = "EXTERN\t".$$line; 1091 $self->{value} .= ":NEAR" if ($masm); 1092 last; 1093 }; 1094 /\.globl|.global/ 1095 && do { $self->{value} = $masm?"PUBLIC":"global"; 1096 $self->{value} .= "\t".$$line; 1097 last; 1098 }; 1099 /\.size/ && do { if (defined($current_function)) { 1100 undef $self->{value}; 1101 if ($current_function->{abi} eq "svr4") { 1102 $self->{value}="${decor}SEH_end_$current_function->{name}:"; 1103 $self->{value}.=":\n" if($masm); 1104 } 1105 $self->{value}.="$current_function->{name}\tENDP" if($masm && $current_function->{name}); 1106 undef $current_function; 1107 } 1108 last; 1109 }; 1110 /\.align/ && do { my $max = ($masm && $masm>=$masmref) ? 256 : 4096; 1111 $self->{value} = "ALIGN\t".($$line>$max?$max:$$line); 1112 last; 1113 }; 1114 /\.(value|long|rva|quad)/ 1115 && do { my $sz = substr($1,0,1); 1116 my @arr = split(/,\s*/,$$line); 1117 my $last = pop(@arr); 1118 my $conv = sub { my $var=shift; 1119 $var=~s/^(0b[0-1]+)/oct($1)/eig; 1120 $var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm); 1121 if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva")) 1122 { $var=~s/^([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; } 1123 $var; 1124 }; 1125 1126 $sz =~ tr/bvlrq/BWDDQ/; 1127 $self->{value} = "\tD$sz\t"; 1128 for (@arr) { $self->{value} .= &$conv($_).","; } 1129 $self->{value} .= &$conv($last); 1130 last; 1131 }; 1132 /\.byte/ && do { my @str=split(/,\s*/,$$line); 1133 map(s/(0b[0-1]+)/oct($1)/eig,@str); 1134 map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm); 1135 while ($#str>15) { 1136 $self->{value}.="DB\t" 1137 .join(",",@str[0..15])."\n"; 1138 foreach (0..15) { shift @str; } 1139 } 1140 $self->{value}.="DB\t" 1141 .join(",",@str) if (@str); 1142 last; 1143 }; 1144 /\.comm/ && do { my @str=split(/,\s*/,$$line); 1145 my $v=undef; 1146 if ($nasm) { 1147 $v.="common $prefix@str[0] @str[1]"; 1148 } else { 1149 $current_segment = pop(@segment_stack);; 1150 $v="$current_segment\tENDS\n" if ($current_segment); 1151 $current_segment = "_DATA"; 1152 push(@segment_stack, $current_segment); 1153 $v.="$current_segment\tSEGMENT\n"; 1154 $v.="COMM @str[0]:DWORD:".@str[1]/4; 1155 } 1156 $self->{value} = $v; 1157 last; 1158 }; 1159 /^.previous/ && do { 1160 my $v=undef; 1161 if ($nasm) { 1162 pop(@segment_stack); # pop ourselves, we don't need to emit END directive 1163 # pop section so we can emit proper .section name. 1164 $current_segment = pop(@segment_stack); 1165 $v="section $current_segment"; 1166 # Hack again: 1167 # push section/segment to stack. The .previous is currently paired 1168 # with .rodata only. We have to keep extra '.text' on stack for 1169 # situation where there is for example .pdata section 'terminated' 1170 # by new '.text' section. 1171 # 1172 push(@segment_stack, $current_segment); 1173 } else { 1174 $current_segment = pop(@segment_stack); 1175 $v="$current_segment\tENDS\n" if ($current_segment); 1176 $current_segment = pop(@segment_stack); 1177 if ($current_segment =~ /\.text\$/) { 1178 $v.="$current_segment\tSEGMENT "; 1179 $v.=$masm>=$masmref ? "ALIGN(256)" : "PAGE"; 1180 $v.=" 'CODE'"; 1181 push(@segment_stack, $current_segment); 1182 } 1183 } 1184 $self->{value} = $v; 1185 last; 1186 }; 1187 } 1188 $$line = ""; 1189 } 1190 1191 $ret; 1192 } 1193 sub out { 1194 my $self = shift; 1195 $self->{value}; 1196 } 1197} 1198 1199# Upon initial x86_64 introduction SSE>2 extensions were not introduced 1200# yet. In order not to be bothered by tracing exact assembler versions, 1201# but at the same time to provide a bare security minimum of AES-NI, we 1202# hard-code some instructions. Extensions past AES-NI on the other hand 1203# are traced by examining assembler version in individual perlasm 1204# modules... 1205 1206my %regrm = ( "%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3, 1207 "%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7 ); 1208 1209sub rex { 1210 my $opcode=shift; 1211 my ($dst,$src,$rex)=@_; 1212 1213 $rex|=0x04 if($dst>=8); 1214 $rex|=0x01 if($src>=8); 1215 push @$opcode,($rex|0x40) if ($rex); 1216} 1217 1218my $movq = sub { # elderly gas can't handle inter-register movq 1219 my $arg = shift; 1220 my @opcode=(0x66); 1221 if ($arg =~ /%xmm([0-9]+),\s*%r(\w+)/) { 1222 my ($src,$dst)=($1,$2); 1223 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 1224 rex(\@opcode,$src,$dst,0x8); 1225 push @opcode,0x0f,0x7e; 1226 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 1227 @opcode; 1228 } elsif ($arg =~ /%r(\w+),\s*%xmm([0-9]+)/) { 1229 my ($src,$dst)=($2,$1); 1230 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 1231 rex(\@opcode,$src,$dst,0x8); 1232 push @opcode,0x0f,0x6e; 1233 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 1234 @opcode; 1235 } else { 1236 (); 1237 } 1238}; 1239 1240my $pextrd = sub { 1241 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*(%\w+)/) { 1242 my @opcode=(0x66); 1243 my $imm=$1; 1244 my $src=$2; 1245 my $dst=$3; 1246 if ($dst =~ /%r([0-9]+)d/) { $dst = $1; } 1247 elsif ($dst =~ /%e/) { $dst = $regrm{$dst}; } 1248 rex(\@opcode,$src,$dst); 1249 push @opcode,0x0f,0x3a,0x16; 1250 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 1251 push @opcode,$imm; 1252 @opcode; 1253 } else { 1254 (); 1255 } 1256}; 1257 1258my $pinsrd = sub { 1259 if (shift =~ /\$([0-9]+),\s*(%\w+),\s*%xmm([0-9]+)/) { 1260 my @opcode=(0x66); 1261 my $imm=$1; 1262 my $src=$2; 1263 my $dst=$3; 1264 if ($src =~ /%r([0-9]+)/) { $src = $1; } 1265 elsif ($src =~ /%e/) { $src = $regrm{$src}; } 1266 rex(\@opcode,$dst,$src); 1267 push @opcode,0x0f,0x3a,0x22; 1268 push @opcode,0xc0|(($dst&7)<<3)|($src&7); # ModR/M 1269 push @opcode,$imm; 1270 @opcode; 1271 } else { 1272 (); 1273 } 1274}; 1275 1276my $pshufb = sub { 1277 if (shift =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) { 1278 my @opcode=(0x66); 1279 rex(\@opcode,$2,$1); 1280 push @opcode,0x0f,0x38,0x00; 1281 push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M 1282 @opcode; 1283 } else { 1284 (); 1285 } 1286}; 1287 1288my $palignr = sub { 1289 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 1290 my @opcode=(0x66); 1291 rex(\@opcode,$3,$2); 1292 push @opcode,0x0f,0x3a,0x0f; 1293 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 1294 push @opcode,$1; 1295 @opcode; 1296 } else { 1297 (); 1298 } 1299}; 1300 1301my $pclmulqdq = sub { 1302 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 1303 my @opcode=(0x66); 1304 rex(\@opcode,$3,$2); 1305 push @opcode,0x0f,0x3a,0x44; 1306 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 1307 my $c=$1; 1308 push @opcode,$c=~/^0/?oct($c):$c; 1309 @opcode; 1310 } else { 1311 (); 1312 } 1313}; 1314 1315my $rdrand = sub { 1316 if (shift =~ /%[er](\w+)/) { 1317 my @opcode=(); 1318 my $dst=$1; 1319 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 1320 rex(\@opcode,0,$dst,8); 1321 push @opcode,0x0f,0xc7,0xf0|($dst&7); 1322 @opcode; 1323 } else { 1324 (); 1325 } 1326}; 1327 1328my $rdseed = sub { 1329 if (shift =~ /%[er](\w+)/) { 1330 my @opcode=(); 1331 my $dst=$1; 1332 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 1333 rex(\@opcode,0,$dst,8); 1334 push @opcode,0x0f,0xc7,0xf8|($dst&7); 1335 @opcode; 1336 } else { 1337 (); 1338 } 1339}; 1340 1341# Not all AVX-capable assemblers recognize AMD XOP extension. Since we 1342# are using only two instructions hand-code them in order to be excused 1343# from chasing assembler versions... 1344 1345sub rxb { 1346 my $opcode=shift; 1347 my ($dst,$src1,$src2,$rxb)=@_; 1348 1349 $rxb|=0x7<<5; 1350 $rxb&=~(0x04<<5) if($dst>=8); 1351 $rxb&=~(0x01<<5) if($src1>=8); 1352 $rxb&=~(0x02<<5) if($src2>=8); 1353 push @$opcode,$rxb; 1354} 1355 1356my $vprotd = sub { 1357 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 1358 my @opcode=(0x8f); 1359 rxb(\@opcode,$3,$2,-1,0x08); 1360 push @opcode,0x78,0xc2; 1361 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 1362 my $c=$1; 1363 push @opcode,$c=~/^0/?oct($c):$c; 1364 @opcode; 1365 } else { 1366 (); 1367 } 1368}; 1369 1370my $vprotq = sub { 1371 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 1372 my @opcode=(0x8f); 1373 rxb(\@opcode,$3,$2,-1,0x08); 1374 push @opcode,0x78,0xc3; 1375 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 1376 my $c=$1; 1377 push @opcode,$c=~/^0/?oct($c):$c; 1378 @opcode; 1379 } else { 1380 (); 1381 } 1382}; 1383 1384# Intel Control-flow Enforcement Technology extension. All functions and 1385# indirect branch targets will have to start with this instruction... 1386 1387my $endbranch = sub { 1388 (0xf3,0x0f,0x1e,0xfa); 1389}; 1390 1391######################################################################## 1392 1393if ($nasm) { 1394 print <<___; 1395default rel 1396%define XMMWORD 1397%define YMMWORD 1398%define ZMMWORD 1399___ 1400} elsif ($masm) { 1401 print <<___; 1402OPTION DOTNAME 1403___ 1404} 1405while(defined(my $line=<>)) { 1406 1407 $line =~ s|\R$||; # Better chomp 1408 1409 $line =~ s|[#!].*$||; # get rid of asm-style comments... 1410 $line =~ s|/\*.*\*/||; # ... and C-style comments... 1411 $line =~ s|^\s+||; # ... and skip whitespaces in beginning 1412 $line =~ s|\s+$||; # ... and at the end 1413 1414 if (my $label=label->re(\$line)) { print $label->out(); } 1415 1416 if (my $directive=directive->re(\$line)) { 1417 printf "%s",$directive->out(); 1418 } else { 1419 if (my $vex_prefix=vex_prefix->re(\$line)) { 1420 printf "%s",$vex_prefix->out(); 1421 } 1422 if (my $opcode=opcode->re(\$line)) { 1423 my $asm = eval("\$".$opcode->mnemonic()); 1424 1425 if ((ref($asm) eq 'CODE') && scalar(my @bytes=&$asm($line))) { 1426 print $gas?".byte\t":"DB\t",join(',',@bytes),"\n"; 1427 next; 1428 } 1429 1430 my @args; 1431 ARGUMENT: while (1) { 1432 my $arg; 1433 1434 ($arg=register->re(\$line, $opcode))|| 1435 ($arg=const->re(\$line)) || 1436 ($arg=ea->re(\$line, $opcode)) || 1437 ($arg=expr->re(\$line, $opcode)) || 1438 last ARGUMENT; 1439 1440 push @args,$arg; 1441 1442 last ARGUMENT if ($line !~ /^,/); 1443 1444 $line =~ s/^,\s*//; 1445 } # ARGUMENT: 1446 1447 if ($#args>=0) { 1448 my $insn; 1449 my $sz=$opcode->size(); 1450 1451 if ($gas) { 1452 $insn = $opcode->out($#args>=1?$args[$#args]->size():$sz); 1453 @args = map($_->out($sz),@args); 1454 printf "\t%s\t%s",$insn,join(",",@args); 1455 } else { 1456 $insn = $opcode->out(); 1457 foreach (@args) { 1458 my $arg = $_->out(); 1459 # $insn.=$sz compensates for movq, pinsrw, ... 1460 if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; } 1461 if ($arg =~ /^ymm[0-9]+$/) { $insn.=$sz; $sz="y" if(!$sz); last; } 1462 if ($arg =~ /^zmm[0-9]+$/) { $insn.=$sz; $sz="z" if(!$sz); last; } 1463 if ($arg =~ /^mm[0-9]+$/) { $insn.=$sz; $sz="q" if(!$sz); last; } 1464 } 1465 @args = reverse(@args); 1466 undef $sz if ($nasm && $opcode->mnemonic() eq "lea"); 1467 printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args)); 1468 } 1469 } else { 1470 printf "\t%s",$opcode->out(); 1471 } 1472 } 1473 } 1474 1475 print $line,"\n"; 1476} 1477 1478print "$cet_property" if ($cet_property); 1479print "\n$current_segment\tENDS\n" if ($current_segment && $masm); 1480print "END\n" if ($masm); 1481 1482close STDOUT or die "error closing STDOUT: $!;" 1483 1484 ################################################# 1485# Cross-reference x86_64 ABI "card" 1486# 1487# Unix Win64 1488# %rax * * 1489# %rbx - - 1490# %rcx #4 #1 1491# %rdx #3 #2 1492# %rsi #2 - 1493# %rdi #1 - 1494# %rbp - - 1495# %rsp - - 1496# %r8 #5 #3 1497# %r9 #6 #4 1498# %r10 * * 1499# %r11 * * 1500# %r12 - - 1501# %r13 - - 1502# %r14 - - 1503# %r15 - - 1504# 1505# (*) volatile register 1506# (-) preserved by callee 1507# (#) Nth argument, volatile 1508# 1509# In Unix terms top of stack is argument transfer area for arguments 1510# which could not be accommodated in registers. Or in other words 7th 1511# [integer] argument resides at 8(%rsp) upon function entry point. 1512# 128 bytes above %rsp constitute a "red zone" which is not touched 1513# by signal handlers and can be used as temporal storage without 1514# allocating a frame. 1515# 1516# In Win64 terms N*8 bytes on top of stack is argument transfer area, 1517# which belongs to/can be overwritten by callee. N is the number of 1518# arguments passed to callee, *but* not less than 4! This means that 1519# upon function entry point 5th argument resides at 40(%rsp), as well 1520# as that 32 bytes from 8(%rsp) can always be used as temporal 1521# storage [without allocating a frame]. One can actually argue that 1522# one can assume a "red zone" above stack pointer under Win64 as well. 1523# Point is that at apparently no occasion Windows kernel would alter 1524# the area above user stack pointer in true asynchronous manner... 1525# 1526# All the above means that if assembler programmer adheres to Unix 1527# register and stack layout, but disregards the "red zone" existence, 1528# it's possible to use following prologue and epilogue to "gear" from 1529# Unix to Win64 ABI in leaf functions with not more than 6 arguments. 1530# 1531# omnipotent_function: 1532# ifdef WIN64 1533# movq %rdi,8(%rsp) 1534# movq %rsi,16(%rsp) 1535# movq %rcx,%rdi ; if 1st argument is actually present 1536# movq %rdx,%rsi ; if 2nd argument is actually ... 1537# movq %r8,%rdx ; if 3rd argument is ... 1538# movq %r9,%rcx ; if 4th argument ... 1539# movq 40(%rsp),%r8 ; if 5th ... 1540# movq 48(%rsp),%r9 ; if 6th ... 1541# endif 1542# ... 1543# ifdef WIN64 1544# movq 8(%rsp),%rdi 1545# movq 16(%rsp),%rsi 1546# endif 1547# ret 1548# 1549 ################################################# 1550# Win64 SEH, Structured Exception Handling. 1551# 1552# Unlike on Unix systems(*) lack of Win64 stack unwinding information 1553# has undesired side-effect at run-time: if an exception is raised in 1554# assembler subroutine such as those in question (basically we're 1555# referring to segmentation violations caused by malformed input 1556# parameters), the application is briskly terminated without invoking 1557# any exception handlers, most notably without generating memory dump 1558# or any user notification whatsoever. This poses a problem. It's 1559# possible to address it by registering custom language-specific 1560# handler that would restore processor context to the state at 1561# subroutine entry point and return "exception is not handled, keep 1562# unwinding" code. Writing such handler can be a challenge... But it's 1563# doable, though requires certain coding convention. Consider following 1564# snippet: 1565# 1566# .type function,@function 1567# function: 1568# movq %rsp,%rax # copy rsp to volatile register 1569# pushq %r15 # save non-volatile registers 1570# pushq %rbx 1571# pushq %rbp 1572# movq %rsp,%r11 1573# subq %rdi,%r11 # prepare [variable] stack frame 1574# andq $-64,%r11 1575# movq %rax,0(%r11) # check for exceptions 1576# movq %r11,%rsp # allocate [variable] stack frame 1577# movq %rax,0(%rsp) # save original rsp value 1578# magic_point: 1579# ... 1580# movq 0(%rsp),%rcx # pull original rsp value 1581# movq -24(%rcx),%rbp # restore non-volatile registers 1582# movq -16(%rcx),%rbx 1583# movq -8(%rcx),%r15 1584# movq %rcx,%rsp # restore original rsp 1585# magic_epilogue: 1586# ret 1587# .size function,.-function 1588# 1589# The key is that up to magic_point copy of original rsp value remains 1590# in chosen volatile register and no non-volatile register, except for 1591# rsp, is modified. While past magic_point rsp remains constant till 1592# the very end of the function. In this case custom language-specific 1593# exception handler would look like this: 1594# 1595# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, 1596# CONTEXT *context,DISPATCHER_CONTEXT *disp) 1597# { ULONG64 *rsp = (ULONG64 *)context->Rax; 1598# ULONG64 rip = context->Rip; 1599# 1600# if (rip >= magic_point) 1601# { rsp = (ULONG64 *)context->Rsp; 1602# if (rip < magic_epilogue) 1603# { rsp = (ULONG64 *)rsp[0]; 1604# context->Rbp = rsp[-3]; 1605# context->Rbx = rsp[-2]; 1606# context->R15 = rsp[-1]; 1607# } 1608# } 1609# context->Rsp = (ULONG64)rsp; 1610# context->Rdi = rsp[1]; 1611# context->Rsi = rsp[2]; 1612# 1613# memcpy (disp->ContextRecord,context,sizeof(CONTEXT)); 1614# RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase, 1615# dips->ControlPc,disp->FunctionEntry,disp->ContextRecord, 1616# &disp->HandlerData,&disp->EstablisherFrame,NULL); 1617# return ExceptionContinueSearch; 1618# } 1619# 1620# It's appropriate to implement this handler in assembler, directly in 1621# function's module. In order to do that one has to know members' 1622# offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant 1623# values. Here they are: 1624# 1625# CONTEXT.Rax 120 1626# CONTEXT.Rcx 128 1627# CONTEXT.Rdx 136 1628# CONTEXT.Rbx 144 1629# CONTEXT.Rsp 152 1630# CONTEXT.Rbp 160 1631# CONTEXT.Rsi 168 1632# CONTEXT.Rdi 176 1633# CONTEXT.R8 184 1634# CONTEXT.R9 192 1635# CONTEXT.R10 200 1636# CONTEXT.R11 208 1637# CONTEXT.R12 216 1638# CONTEXT.R13 224 1639# CONTEXT.R14 232 1640# CONTEXT.R15 240 1641# CONTEXT.Rip 248 1642# CONTEXT.Xmm6 512 1643# sizeof(CONTEXT) 1232 1644# DISPATCHER_CONTEXT.ControlPc 0 1645# DISPATCHER_CONTEXT.ImageBase 8 1646# DISPATCHER_CONTEXT.FunctionEntry 16 1647# DISPATCHER_CONTEXT.EstablisherFrame 24 1648# DISPATCHER_CONTEXT.TargetIp 32 1649# DISPATCHER_CONTEXT.ContextRecord 40 1650# DISPATCHER_CONTEXT.LanguageHandler 48 1651# DISPATCHER_CONTEXT.HandlerData 56 1652# UNW_FLAG_NHANDLER 0 1653# ExceptionContinueSearch 1 1654# 1655# In order to tie the handler to the function one has to compose 1656# couple of structures: one for .xdata segment and one for .pdata. 1657# 1658# UNWIND_INFO structure for .xdata segment would be 1659# 1660# function_unwind_info: 1661# .byte 9,0,0,0 1662# .rva handler 1663# 1664# This structure designates exception handler for a function with 1665# zero-length prologue, no stack frame or frame register. 1666# 1667# To facilitate composing of .pdata structures, auto-generated "gear" 1668# prologue copies rsp value to rax and denotes next instruction with 1669# .LSEH_begin_{function_name} label. This essentially defines the SEH 1670# styling rule mentioned in the beginning. Position of this label is 1671# chosen in such manner that possible exceptions raised in the "gear" 1672# prologue would be accounted to caller and unwound from latter's frame. 1673# End of function is marked with respective .LSEH_end_{function_name} 1674# label. To summarize, .pdata segment would contain 1675# 1676# .rva .LSEH_begin_function 1677# .rva .LSEH_end_function 1678# .rva function_unwind_info 1679# 1680# Reference to function_unwind_info from .xdata segment is the anchor. 1681# In case you wonder why references are 32-bit .rvas and not 64-bit 1682# .quads. References put into these two segments are required to be 1683# *relative* to the base address of the current binary module, a.k.a. 1684# image base. No Win64 module, be it .exe or .dll, can be larger than 1685# 2GB and thus such relative references can be and are accommodated in 1686# 32 bits. 1687# 1688# Having reviewed the example function code, one can argue that "movq 1689# %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix 1690# rax would contain an undefined value. If this "offends" you, use 1691# another register and refrain from modifying rax till magic_point is 1692# reached, i.e. as if it was a non-volatile register. If more registers 1693# are required prior [variable] frame setup is completed, note that 1694# nobody says that you can have only one "magic point." You can 1695# "liberate" non-volatile registers by denoting last stack off-load 1696# instruction and reflecting it in finer grade unwind logic in handler. 1697# After all, isn't it why it's called *language-specific* handler... 1698# 1699# SE handlers are also involved in unwinding stack when executable is 1700# profiled or debugged. Profiling implies additional limitations that 1701# are too subtle to discuss here. For now it's sufficient to say that 1702# in order to simplify handlers one should either a) offload original 1703# %rsp to stack (like discussed above); or b) if you have a register to 1704# spare for frame pointer, choose volatile one. 1705# 1706# (*) Note that we're talking about run-time, not debug-time. Lack of 1707# unwind information makes debugging hard on both Windows and 1708# Unix. "Unlike" refers to the fact that on Unix signal handler 1709# will always be invoked, core dumped and appropriate exit code 1710# returned to parent (for user notification). 1711