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# 7. Stick to explicit ip-relative addressing. If you have to use 51# GOTPCREL addressing, stick to mov symbol@GOTPCREL(%rip),%r??. 52# Both are recognized and translated to proper Win64 addressing 53# modes. 54# 55# 8. In order to provide for structured exception handling unified 56# Win64 prologue copies %rsp value to %rax. For further details 57# see SEH paragraph at the end. 58# 9. .init segment is allowed to contain calls to functions only. 59# a. If function accepts more than 4 arguments *and* >4th argument 60# is declared as non 64-bit value, do clear its upper part. 61# 62# TODO(https://crbug.com/boringssl/259): The dual-ABI mechanism described here 63# does not quite unwind correctly on Windows. The seh_directive logic below has 64# the start of a new mechanism. 65 66 67use strict; 68 69my $flavour = shift; 70my $output = shift; 71if ($flavour =~ /\./) { $output = $flavour; undef $flavour; } 72 73open STDOUT,">$output" || die "can't open $output: $!" 74 if (defined($output)); 75 76my $gas=1; $gas=0 if ($output =~ /\.asm$/); 77my $elf=1; $elf=0 if (!$gas); 78my $apple=0; 79my $win64=0; 80my $prefix=""; 81my $decor=".L"; 82 83my $masmref=8 + 50727*2**-32; # 8.00.50727 shipped with VS2005 84my $masm=0; 85my $PTR=" PTR"; 86 87my $nasmref=2.03; 88my $nasm=0; 89 90if ($flavour eq "mingw64") { $gas=1; $elf=0; $win64=1; 91 # TODO(davidben): Before supporting the 92 # mingw64 perlasm flavour, do away with this 93 # environment variable check. 94 die "mingw64 not supported"; 95 $prefix=`echo __USER_LABEL_PREFIX__ | $ENV{CC} -E -P -`; 96 $prefix =~ s|\R$||; # Better chomp 97 } 98elsif ($flavour eq "macosx") { $gas=1; $elf=0; $apple=1; $prefix="_"; $decor="L\$"; } 99elsif ($flavour eq "masm") { $gas=0; $elf=0; $masm=$masmref; $win64=1; $decor="\$L\$"; } 100elsif ($flavour eq "nasm") { $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $decor="\$L\$"; $PTR=""; } 101elsif (!$gas) { die "unknown flavour $flavour"; } 102 103my $current_segment; 104my $current_function; 105my %globals; 106 107{ package opcode; # pick up opcodes 108 sub re { 109 my ($class, $line) = @_; 110 my $self = {}; 111 my $ret; 112 113 if ($$line =~ /^([a-z][a-z0-9]*)/i) { 114 bless $self,$class; 115 $self->{op} = $1; 116 $ret = $self; 117 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 118 119 undef $self->{sz}; 120 if ($self->{op} =~ /^(movz)x?([bw]).*/) { # movz is pain... 121 $self->{op} = $1; 122 $self->{sz} = $2; 123 } elsif ($self->{op} =~ /call|jmp/) { 124 $self->{sz} = ""; 125 } elsif ($self->{op} =~ /^p/ && $' !~ /^(ush|op|insrw)/) { # SSEn 126 $self->{sz} = ""; 127 } elsif ($self->{op} =~ /^[vk]/) { # VEX or k* such as kmov 128 $self->{sz} = ""; 129 } elsif ($self->{op} =~ /mov[dq]/ && $$line =~ /%xmm/) { 130 $self->{sz} = ""; 131 } elsif ($self->{op} =~ /^or([qlwb])$/) { 132 $self->{op} = "or"; 133 $self->{sz} = $1; 134 } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) { 135 $self->{op} = $1; 136 $self->{sz} = $2; 137 } 138 } 139 $ret; 140 } 141 sub size { 142 my ($self, $sz) = @_; 143 $self->{sz} = $sz if (defined($sz) && !defined($self->{sz})); 144 $self->{sz}; 145 } 146 sub out { 147 my $self = shift; 148 if ($gas) { 149 if ($self->{op} eq "movz") { # movz is pain... 150 sprintf "%s%s%s",$self->{op},$self->{sz},shift; 151 } elsif ($self->{op} =~ /^set/) { 152 "$self->{op}"; 153 } elsif ($self->{op} eq "ret") { 154 my $epilogue = ""; 155 if ($win64 && $current_function->{abi} eq "svr4") { 156 $epilogue = "movq 8(%rsp),%rdi\n\t" . 157 "movq 16(%rsp),%rsi\n\t"; 158 } 159 $epilogue . "ret"; 160 } elsif ($self->{op} eq "call" && !$elf && $current_segment eq ".init") { 161 ".p2align\t3\n\t.quad"; 162 } else { 163 "$self->{op}$self->{sz}"; 164 } 165 } else { 166 $self->{op} =~ s/^movz/movzx/; 167 if ($self->{op} eq "ret") { 168 $self->{op} = ""; 169 if ($win64 && $current_function->{abi} eq "svr4") { 170 $self->{op} = "mov rdi,QWORD$PTR\[8+rsp\]\t;WIN64 epilogue\n\t". 171 "mov rsi,QWORD$PTR\[16+rsp\]\n\t"; 172 } 173 $self->{op} .= "ret"; 174 } elsif ($self->{op} =~ /^(pop|push)f/) { 175 $self->{op} .= $self->{sz}; 176 } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") { 177 $self->{op} = "\tDQ"; 178 } 179 $self->{op}; 180 } 181 } 182 sub mnemonic { 183 my ($self, $op) = @_; 184 $self->{op}=$op if (defined($op)); 185 $self->{op}; 186 } 187} 188{ package const; # pick up constants, which start with $ 189 sub re { 190 my ($class, $line) = @_; 191 my $self = {}; 192 my $ret; 193 194 if ($$line =~ /^\$([^,]+)/) { 195 bless $self, $class; 196 $self->{value} = $1; 197 $ret = $self; 198 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 199 } 200 $ret; 201 } 202 sub out { 203 my $self = shift; 204 205 $self->{value} =~ s/\b(0b[0-1]+)/oct($1)/eig; 206 if ($gas) { 207 # Solaris /usr/ccs/bin/as can't handle multiplications 208 # in $self->{value} 209 my $value = $self->{value}; 210 no warnings; # oct might complain about overflow, ignore here... 211 $value =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi; 212 if ($value =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg) { 213 $self->{value} = $value; 214 } 215 sprintf "\$%s",$self->{value}; 216 } else { 217 my $value = $self->{value}; 218 $value =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm); 219 sprintf "%s",$value; 220 } 221 } 222} 223{ package ea; # pick up effective addresses: expr(%reg,%reg,scale) 224 225 my %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR", 226 l=>"DWORD$PTR", d=>"DWORD$PTR", 227 q=>"QWORD$PTR", o=>"OWORD$PTR", 228 x=>"XMMWORD$PTR", y=>"YMMWORD$PTR", 229 z=>"ZMMWORD$PTR" ) if (!$gas); 230 231 sub re { 232 my ($class, $line, $opcode) = @_; 233 my $self = {}; 234 my $ret; 235 236 # optional * ----vvv--- appears in indirect jmp/call 237 if ($$line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)((?:{[^}]+})*)/) { 238 bless $self, $class; 239 $self->{asterisk} = $1; 240 $self->{label} = $2; 241 ($self->{base},$self->{index},$self->{scale})=split(/,/,$3); 242 $self->{scale} = 1 if (!defined($self->{scale})); 243 $self->{opmask} = $4; 244 $ret = $self; 245 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 246 247 if ($win64 && $self->{label} =~ s/\@GOTPCREL//) { 248 die if ($opcode->mnemonic() ne "mov"); 249 $opcode->mnemonic("lea"); 250 } 251 $self->{base} =~ s/^%//; 252 $self->{index} =~ s/^%// if (defined($self->{index})); 253 $self->{opcode} = $opcode; 254 } 255 $ret; 256 } 257 sub size {} 258 sub out { 259 my ($self, $sz) = @_; 260 261 $self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 262 $self->{label} =~ s/\.L/$decor/g; 263 264 # Silently convert all EAs to 64-bit. This is required for 265 # elder GNU assembler and results in more compact code, 266 # *but* most importantly AES module depends on this feature! 267 $self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; 268 $self->{base} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; 269 270 # Solaris /usr/ccs/bin/as can't handle multiplications 271 # in $self->{label}... 272 use integer; 273 $self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi; 274 $self->{label} =~ s/\b([0-9]+\s*[\*\/\%]\s*[0-9]+)\b/eval($1)/eg; 275 276 # Some assemblers insist on signed presentation of 32-bit 277 # offsets, but sign extension is a tricky business in perl... 278 if ((1<<31)<<1) { 279 $self->{label} =~ s/\b([0-9]+)\b/$1<<32>>32/eg; 280 } else { 281 $self->{label} =~ s/\b([0-9]+)\b/$1>>0/eg; 282 } 283 284 # if base register is %rbp or %r13, see if it's possible to 285 # flip base and index registers [for better performance] 286 if (!$self->{label} && $self->{index} && $self->{scale}==1 && 287 $self->{base} =~ /(rbp|r13)/) { 288 $self->{base} = $self->{index}; $self->{index} = $1; 289 } 290 291 if ($gas) { 292 $self->{label} =~ s/^___imp_/__imp__/ if ($flavour eq "mingw64"); 293 294 if (defined($self->{index})) { 295 sprintf "%s%s(%s,%%%s,%d)%s", 296 $self->{asterisk},$self->{label}, 297 $self->{base}?"%$self->{base}":"", 298 $self->{index},$self->{scale}, 299 $self->{opmask}; 300 } else { 301 sprintf "%s%s(%%%s)%s", $self->{asterisk},$self->{label}, 302 $self->{base},$self->{opmask}; 303 } 304 } else { 305 $self->{label} =~ s/\./\$/g; 306 $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig; 307 $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/); 308 309 my $mnemonic = $self->{opcode}->mnemonic(); 310 ($self->{asterisk}) && ($sz="q") || 311 ($mnemonic =~ /^v?mov([qd])$/) && ($sz=$1) || 312 ($mnemonic =~ /^v?pinsr([qdwb])$/) && ($sz=$1) || 313 ($mnemonic =~ /^vpbroadcast([qdwb])$/) && ($sz=$1) || 314 ($mnemonic =~ /^v(?!perm)[a-z]+[fi]128$/) && ($sz="x"); 315 316 $self->{opmask} =~ s/%(k[0-7])/$1/; 317 318 if (defined($self->{index})) { 319 sprintf "%s[%s%s*%d%s]%s",$szmap{$sz}, 320 $self->{label}?"$self->{label}+":"", 321 $self->{index},$self->{scale}, 322 $self->{base}?"+$self->{base}":"", 323 $self->{opmask}; 324 } elsif ($self->{base} eq "rip") { 325 sprintf "%s[%s]",$szmap{$sz},$self->{label}; 326 } else { 327 sprintf "%s[%s%s]%s", $szmap{$sz}, 328 $self->{label}?"$self->{label}+":"", 329 $self->{base},$self->{opmask}; 330 } 331 } 332 } 333} 334{ package register; # pick up registers, which start with %. 335 sub re { 336 my ($class, $line, $opcode) = @_; 337 my $self = {}; 338 my $ret; 339 340 # optional * ----vvv--- appears in indirect jmp/call 341 if ($$line =~ /^(\*?)%(\w+)((?:{[^}]+})*)/) { 342 bless $self,$class; 343 $self->{asterisk} = $1; 344 $self->{value} = $2; 345 $self->{opmask} = $3; 346 $opcode->size($self->size()); 347 $ret = $self; 348 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 349 } 350 $ret; 351 } 352 sub size { 353 my $self = shift; 354 my $ret; 355 356 if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; } 357 elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; } 358 elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; } 359 elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; } 360 elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; } 361 elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; } 362 elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; } 363 elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; } 364 365 $ret; 366 } 367 sub out { 368 my $self = shift; 369 if ($gas) { sprintf "%s%%%s%s", $self->{asterisk}, 370 $self->{value}, 371 $self->{opmask}; } 372 else { $self->{opmask} =~ s/%(k[0-7])/$1/; 373 $self->{value}.$self->{opmask}; } 374 } 375} 376{ package label; # pick up labels, which end with : 377 sub re { 378 my ($class, $line) = @_; 379 my $self = {}; 380 my $ret; 381 382 if ($$line =~ /(^[\.\w]+)\:/) { 383 bless $self,$class; 384 $self->{value} = $1; 385 $ret = $self; 386 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 387 388 $self->{value} =~ s/^\.L/$decor/; 389 } 390 $ret; 391 } 392 sub out { 393 my $self = shift; 394 395 if ($gas) { 396 my $func = ($globals{$self->{value}} or $self->{value}) . ":"; 397 if ($win64 && $current_function->{name} eq $self->{value} 398 && $current_function->{abi} eq "svr4") { 399 $func .= "\n"; 400 $func .= " movq %rdi,8(%rsp)\n"; 401 $func .= " movq %rsi,16(%rsp)\n"; 402 $func .= " movq %rsp,%rax\n"; 403 $func .= "${decor}SEH_begin_$current_function->{name}:\n"; 404 my $narg = $current_function->{narg}; 405 $narg=6 if (!defined($narg)); 406 $func .= " movq %rcx,%rdi\n" if ($narg>0); 407 $func .= " movq %rdx,%rsi\n" if ($narg>1); 408 $func .= " movq %r8,%rdx\n" if ($narg>2); 409 $func .= " movq %r9,%rcx\n" if ($narg>3); 410 $func .= " movq 40(%rsp),%r8\n" if ($narg>4); 411 $func .= " movq 48(%rsp),%r9\n" if ($narg>5); 412 } 413 $func; 414 } elsif ($self->{value} ne "$current_function->{name}") { 415 # Make all labels in masm global. 416 $self->{value} .= ":" if ($masm); 417 $self->{value} . ":"; 418 } elsif ($win64 && $current_function->{abi} eq "svr4") { 419 my $func = "$current_function->{name}" . 420 ($nasm ? ":" : "\tPROC $current_function->{scope}") . 421 "\n"; 422 $func .= " mov QWORD$PTR\[8+rsp\],rdi\t;WIN64 prologue\n"; 423 $func .= " mov QWORD$PTR\[16+rsp\],rsi\n"; 424 $func .= " mov rax,rsp\n"; 425 $func .= "${decor}SEH_begin_$current_function->{name}:"; 426 $func .= ":" if ($masm); 427 $func .= "\n"; 428 my $narg = $current_function->{narg}; 429 $narg=6 if (!defined($narg)); 430 $func .= " mov rdi,rcx\n" if ($narg>0); 431 $func .= " mov rsi,rdx\n" if ($narg>1); 432 $func .= " mov rdx,r8\n" if ($narg>2); 433 $func .= " mov rcx,r9\n" if ($narg>3); 434 $func .= " mov r8,QWORD$PTR\[40+rsp\]\n" if ($narg>4); 435 $func .= " mov r9,QWORD$PTR\[48+rsp\]\n" if ($narg>5); 436 $func .= "\n"; 437 } else { 438 "$current_function->{name}". 439 ($nasm ? ":" : "\tPROC $current_function->{scope}"); 440 } 441 } 442} 443{ package expr; # pick up expressions 444 sub re { 445 my ($class, $line, $opcode) = @_; 446 my $self = {}; 447 my $ret; 448 449 if ($$line =~ /(^[^,]+)/) { 450 bless $self,$class; 451 $self->{value} = $1; 452 $ret = $self; 453 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 454 455 $self->{value} =~ s/\@PLT// if (!$elf); 456 $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 457 $self->{value} =~ s/\.L/$decor/g; 458 $self->{opcode} = $opcode; 459 } 460 $ret; 461 } 462 sub out { 463 my $self = shift; 464 if ($nasm && $self->{opcode}->mnemonic()=~m/^j(?![re]cxz)/) { 465 "NEAR ".$self->{value}; 466 } else { 467 $self->{value}; 468 } 469 } 470} 471{ package cfi_directive; 472 # CFI directives annotate instructions that are significant for 473 # stack unwinding procedure compliant with DWARF specification, 474 # see http://dwarfstd.org/. Besides naturally expected for this 475 # script platform-specific filtering function, this module adds 476 # three auxiliary synthetic directives not recognized by [GNU] 477 # assembler: 478 # 479 # - .cfi_push to annotate push instructions in prologue, which 480 # translates to .cfi_adjust_cfa_offset (if needed) and 481 # .cfi_offset; 482 # - .cfi_pop to annotate pop instructions in epilogue, which 483 # translates to .cfi_adjust_cfa_offset (if needed) and 484 # .cfi_restore; 485 # - [and most notably] .cfi_cfa_expression which encodes 486 # DW_CFA_def_cfa_expression and passes it to .cfi_escape as 487 # byte vector; 488 # 489 # CFA expressions were introduced in DWARF specification version 490 # 3 and describe how to deduce CFA, Canonical Frame Address. This 491 # becomes handy if your stack frame is variable and you can't 492 # spare register for [previous] frame pointer. Suggested directive 493 # syntax is made-up mix of DWARF operator suffixes [subset of] 494 # and references to registers with optional bias. Following example 495 # describes offloaded *original* stack pointer at specific offset 496 # from *current* stack pointer: 497 # 498 # .cfi_cfa_expression %rsp+40,deref,+8 499 # 500 # Final +8 has everything to do with the fact that CFA is defined 501 # as reference to top of caller's stack, and on x86_64 call to 502 # subroutine pushes 8-byte return address. In other words original 503 # stack pointer upon entry to a subroutine is 8 bytes off from CFA. 504 505 # Below constants are taken from "DWARF Expressions" section of the 506 # DWARF specification, section is numbered 7.7 in versions 3 and 4. 507 my %DW_OP_simple = ( # no-arg operators, mapped directly 508 deref => 0x06, dup => 0x12, 509 drop => 0x13, over => 0x14, 510 pick => 0x15, swap => 0x16, 511 rot => 0x17, xderef => 0x18, 512 513 abs => 0x19, and => 0x1a, 514 div => 0x1b, minus => 0x1c, 515 mod => 0x1d, mul => 0x1e, 516 neg => 0x1f, not => 0x20, 517 or => 0x21, plus => 0x22, 518 shl => 0x24, shr => 0x25, 519 shra => 0x26, xor => 0x27, 520 ); 521 522 my %DW_OP_complex = ( # used in specific subroutines 523 constu => 0x10, # uleb128 524 consts => 0x11, # sleb128 525 plus_uconst => 0x23, # uleb128 526 lit0 => 0x30, # add 0-31 to opcode 527 reg0 => 0x50, # add 0-31 to opcode 528 breg0 => 0x70, # add 0-31 to opcole, sleb128 529 regx => 0x90, # uleb28 530 fbreg => 0x91, # sleb128 531 bregx => 0x92, # uleb128, sleb128 532 piece => 0x93, # uleb128 533 ); 534 535 # Following constants are defined in x86_64 ABI supplement, for 536 # example available at https://www.uclibc.org/docs/psABI-x86_64.pdf, 537 # see section 3.7 "Stack Unwind Algorithm". 538 my %DW_reg_idx = ( 539 "%rax"=>0, "%rdx"=>1, "%rcx"=>2, "%rbx"=>3, 540 "%rsi"=>4, "%rdi"=>5, "%rbp"=>6, "%rsp"=>7, 541 "%r8" =>8, "%r9" =>9, "%r10"=>10, "%r11"=>11, 542 "%r12"=>12, "%r13"=>13, "%r14"=>14, "%r15"=>15 543 ); 544 545 my ($cfa_reg, $cfa_rsp); 546 my @cfa_stack; 547 548 # [us]leb128 format is variable-length integer representation base 549 # 2^128, with most significant bit of each byte being 0 denoting 550 # *last* most significant digit. See "Variable Length Data" in the 551 # DWARF specification, numbered 7.6 at least in versions 3 and 4. 552 sub sleb128 { 553 use integer; # get right shift extend sign 554 555 my $val = shift; 556 my $sign = ($val < 0) ? -1 : 0; 557 my @ret = (); 558 559 while(1) { 560 push @ret, $val&0x7f; 561 562 # see if remaining bits are same and equal to most 563 # significant bit of the current digit, if so, it's 564 # last digit... 565 last if (($val>>6) == $sign); 566 567 @ret[-1] |= 0x80; 568 $val >>= 7; 569 } 570 571 return @ret; 572 } 573 sub uleb128 { 574 my $val = shift; 575 my @ret = (); 576 577 while(1) { 578 push @ret, $val&0x7f; 579 580 # see if it's last significant digit... 581 last if (($val >>= 7) == 0); 582 583 @ret[-1] |= 0x80; 584 } 585 586 return @ret; 587 } 588 sub const { 589 my $val = shift; 590 591 if ($val >= 0 && $val < 32) { 592 return ($DW_OP_complex{lit0}+$val); 593 } 594 return ($DW_OP_complex{consts}, sleb128($val)); 595 } 596 sub reg { 597 my $val = shift; 598 599 return if ($val !~ m/^(%r\w+)(?:([\+\-])((?:0x)?[0-9a-f]+))?/); 600 601 my $reg = $DW_reg_idx{$1}; 602 my $off = eval ("0 $2 $3"); 603 604 return (($DW_OP_complex{breg0} + $reg), sleb128($off)); 605 # Yes, we use DW_OP_bregX+0 to push register value and not 606 # DW_OP_regX, because latter would require even DW_OP_piece, 607 # which would be a waste under the circumstances. If you have 608 # to use DWP_OP_reg, use "regx:N"... 609 } 610 sub cfa_expression { 611 my $line = shift; 612 my @ret; 613 614 foreach my $token (split(/,\s*/,$line)) { 615 if ($token =~ /^%r/) { 616 push @ret,reg($token); 617 } elsif ($token =~ /((?:0x)?[0-9a-f]+)\((%r\w+)\)/) { 618 push @ret,reg("$2+$1"); 619 } elsif ($token =~ /(\w+):(\-?(?:0x)?[0-9a-f]+)(U?)/i) { 620 my $i = 1*eval($2); 621 push @ret,$DW_OP_complex{$1}, ($3 ? uleb128($i) : sleb128($i)); 622 } elsif (my $i = 1*eval($token) or $token eq "0") { 623 if ($token =~ /^\+/) { 624 push @ret,$DW_OP_complex{plus_uconst},uleb128($i); 625 } else { 626 push @ret,const($i); 627 } 628 } else { 629 push @ret,$DW_OP_simple{$token}; 630 } 631 } 632 633 # Finally we return DW_CFA_def_cfa_expression, 15, followed by 634 # length of the expression and of course the expression itself. 635 return (15,scalar(@ret),@ret); 636 } 637 sub re { 638 my ($class, $line) = @_; 639 my $self = {}; 640 my $ret; 641 642 if ($$line =~ s/^\s*\.cfi_(\w+)\s*//) { 643 bless $self,$class; 644 $ret = $self; 645 undef $self->{value}; 646 my $dir = $1; 647 648 SWITCH: for ($dir) { 649 # What is $cfa_rsp? Effectively it's difference between %rsp 650 # value and current CFA, Canonical Frame Address, which is 651 # why it starts with -8. Recall that CFA is top of caller's 652 # stack... 653 /startproc/ && do { ($cfa_reg, $cfa_rsp) = ("%rsp", -8); last; }; 654 /endproc/ && do { ($cfa_reg, $cfa_rsp) = ("%rsp", 0); last; }; 655 /def_cfa_register/ 656 && do { $cfa_reg = $$line; last; }; 657 /def_cfa_offset/ 658 && do { $cfa_rsp = -1*eval($$line) if ($cfa_reg eq "%rsp"); 659 last; 660 }; 661 /adjust_cfa_offset/ 662 && do { $cfa_rsp -= 1*eval($$line) if ($cfa_reg eq "%rsp"); 663 last; 664 }; 665 /def_cfa/ && do { if ($$line =~ /(%r\w+)\s*,\s*(.+)/) { 666 $cfa_reg = $1; 667 $cfa_rsp = -1*eval($2) if ($cfa_reg eq "%rsp"); 668 } 669 last; 670 }; 671 /push/ && do { $dir = undef; 672 $cfa_rsp -= 8; 673 if ($cfa_reg eq "%rsp") { 674 $self->{value} = ".cfi_adjust_cfa_offset\t8\n"; 675 } 676 $self->{value} .= ".cfi_offset\t$$line,$cfa_rsp"; 677 last; 678 }; 679 /pop/ && do { $dir = undef; 680 $cfa_rsp += 8; 681 if ($cfa_reg eq "%rsp") { 682 $self->{value} = ".cfi_adjust_cfa_offset\t-8\n"; 683 } 684 $self->{value} .= ".cfi_restore\t$$line"; 685 last; 686 }; 687 /cfa_expression/ 688 && do { $dir = undef; 689 $self->{value} = ".cfi_escape\t" . 690 join(",", map(sprintf("0x%02x", $_), 691 cfa_expression($$line))); 692 last; 693 }; 694 /remember_state/ 695 && do { push @cfa_stack, [$cfa_reg, $cfa_rsp]; 696 last; 697 }; 698 /restore_state/ 699 && do { ($cfa_reg, $cfa_rsp) = @{pop @cfa_stack}; 700 last; 701 }; 702 } 703 704 $self->{value} = ".cfi_$dir\t$$line" if ($dir); 705 706 $$line = ""; 707 } 708 709 return $ret; 710 } 711 sub out { 712 my $self = shift; 713 return ($elf ? $self->{value} : undef); 714 } 715} 716{ package seh_directive; 717 # This implements directives, like MASM, gas, and clang-assembler for 718 # specifying Windows unwind codes. See 719 # https://learn.microsoft.com/en-us/cpp/build/exception-handling-x64?view=msvc-170 720 # for details on the Windows unwind mechanism. As perlasm generally uses gas 721 # syntax, the syntax is patterned after the gas spelling, described in 722 # https://sourceware.org/legacy-ml/binutils/2009-08/msg00193.html 723 # 724 # TODO(https://crbug.com/boringssl/571): Translate to the MASM directives 725 # when using the MASM output. Emit as-is when using "mingw64" output, which 726 # is Windows with gas syntax. 727 # 728 # TODO(https://crbug.com/boringssl/259): For now, SEH directives are ignored 729 # on non-Windows platforms. This means functions need to specify both CFI 730 # and SEH directives, often redundantly. Ideally we'd abstract between the 731 # two. E.g., we can synthesize CFI from SEH prologues, but SEH does not 732 # annotate epilogs, so we'd need to combine parts from both. Or we can 733 # restrict ourselves to a subset of CFI and synthesize SEH from CFI. 734 # 735 # Additionally, this only supports @abi-omnipotent functions. It is 736 # incompatible with the automatic calling convention conversion. The main 737 # complication is the current scheme modifies RDI and RSI (non-volatile on 738 # Windows) at the start of the function, and saves them in the parameter 739 # stack area. This can be expressed with .seh_savereg, but .seh_savereg is 740 # only usable late in the prologue. However, unwind information gives enough 741 # information to locate the parameter stack area at any point in the 742 # function, so we can defer conversion or implement other schemes. 743 744 my $UWOP_PUSH_NONVOL = 0; 745 my $UWOP_ALLOC_LARGE = 1; 746 my $UWOP_ALLOC_SMALL = 2; 747 my $UWOP_SET_FPREG = 3; 748 my $UWOP_SAVE_NONVOL = 4; 749 my $UWOP_SAVE_NONVOL_FAR = 5; 750 my $UWOP_SAVE_XMM128 = 8; 751 my $UWOP_SAVE_XMM128_FAR = 9; 752 753 my %UWOP_REG_TO_NUMBER = ("%rax" => 0, "%rcx" => 1, "%rdx" => 2, "%rbx" => 3, 754 "%rsp" => 4, "%rbp" => 5, "%rsi" => 6, "%rdi" => 7, 755 map(("%r$_" => $_), (8..15))); 756 my %UWOP_NUMBER_TO_REG = reverse %UWOP_REG_TO_NUMBER; 757 758 # The contents of the pdata and xdata sections so far. 759 my ($xdata, $pdata) = ("", ""); 760 761 my %info; 762 763 my $next_label = 0; 764 my $current_label_func = ""; 765 766 # _new_unwind_label allocates a new label, unique to the file. 767 sub _new_unwind_label { 768 my ($name) = (@_); 769 # Labels only need to be unique, but to make diffs easier to read, scope 770 # them all under the current function. 771 my $func = $current_function->{name}; 772 if ($func ne $current_label_func) { 773 $current_label_func = $func; 774 $next_label = 0; 775 } 776 777 my $num = $next_label++; 778 return ".LSEH_${name}_${func}_${num}"; 779 } 780 781 sub _check_in_proc { 782 die "Missing .seh_startproc directive" unless %info; 783 } 784 785 sub _check_in_prologue { 786 _check_in_proc(); 787 die "Invalid SEH directive after .seh_endprologue" if defined($info{endprologue}); 788 } 789 790 sub _check_not_in_proc { 791 die "Missing .seh_endproc directive" if %info; 792 } 793 794 sub _startproc { 795 _check_not_in_proc(); 796 if ($current_function->{abi} eq "svr4") { 797 die "SEH directives can only be used with \@abi-omnipotent"; 798 } 799 800 my $info_label = _new_unwind_label("info"); 801 my $start_label = _new_unwind_label("begin"); 802 %info = ( 803 # info_label is the label of the function's entry in .xdata. 804 info_label => $info_label, 805 # start_label is the start of the function. 806 start_label => $start_label, 807 # endprologue is the label of the end of the prologue. 808 endprologue => undef, 809 # unwind_codes contains the textual representation of the 810 # unwind codes in the function so far. 811 unwind_codes => "", 812 # num_codes is the number of 16-bit words in unwind_codes. 813 num_codes => 0, 814 # frame_reg is the number of the frame register, or zero if 815 # there is none. 816 frame_reg => 0, 817 # frame_offset is the offset into the fixed part of the stack that 818 # the frame register points into. 819 frame_offset => 0, 820 # has_offset is whether directives taking an offset have 821 # been used. This is used to check that such directives 822 # come after the fixed portion of the stack frame is established. 823 has_offset => 0, 824 # has_nonpushreg is whether directives other than 825 # .seh_pushreg have been used. This is used to check that 826 # .seh_pushreg directives are first. 827 has_nonpushreg => 0, 828 ); 829 return $start_label; 830 } 831 832 sub _add_unwind_code { 833 my ($op, $value, @extra) = @_; 834 _check_in_prologue(); 835 if ($op != $UWOP_PUSH_NONVOL) { 836 $info{has_nonpushreg} = 1; 837 } elsif ($info{has_nonpushreg}) { 838 die ".seh_pushreg directives must appear first in the prologue"; 839 } 840 841 my $label = _new_unwind_label("prologue"); 842 # Encode an UNWIND_CODE structure. See 843 # https://learn.microsoft.com/en-us/cpp/build/exception-handling-x64?view=msvc-170#struct-unwind_code 844 my $encoded = $op | ($value << 4); 845 my $codes = <<____; 846 .byte $label-$info{start_label} 847 .byte $encoded 848____ 849 # Some opcodes need additional values to encode themselves. 850 foreach (@extra) { 851 $codes .= "\t.value\t$_\n"; 852 } 853 854 $info{num_codes} += 1 + scalar(@extra); 855 # Unwind codes are listed in reverse order. 856 $info{unwind_codes} = $codes . $info{unwind_codes}; 857 return $label; 858 } 859 860 sub _updating_fixed_allocation { 861 _check_in_prologue(); 862 if ($info{frame_reg} != 0) { 863 # Windows documentation does not explicitly forbid .seh_stackalloc 864 # after .seh_setframe, but it appears to have no effect. Offsets are 865 # still relative to the fixed allocation when the frame register was 866 # established. 867 die "fixed allocation may not be increased after .seh_setframe"; 868 } 869 if ($info{has_offset}) { 870 # Windows documentation does not explicitly forbid .seh_savereg 871 # before .seh_stackalloc, but it does not work very well. Offsets 872 # are relative to the top of the final fixed allocation, not where 873 # RSP currently is. 874 die "directives with an offset must come after the fixed allocation is established."; 875 } 876 } 877 878 sub _endproc { 879 _check_in_proc(); 880 if (!defined($info{endprologue})) { 881 die "Missing .seh_endprologue"; 882 } 883 884 my $end_label = _new_unwind_label("end"); 885 # Encode a RUNTIME_FUNCTION. See 886 # https://learn.microsoft.com/en-us/cpp/build/exception-handling-x64?view=msvc-170#struct-runtime_function 887 $pdata .= <<____; 888 .rva $info{start_label} 889 .rva $end_label 890 .rva $info{info_label} 891 892____ 893 894 # Encode an UNWIND_INFO. See 895 # https://learn.microsoft.com/en-us/cpp/build/exception-handling-x64?view=msvc-170#struct-unwind_info 896 my $frame_encoded = $info{frame_reg} | (($info{frame_offset} / 16) << 4); 897 $xdata .= <<____; 898$info{info_label}: 899 .byte 1 # version 1, no flags 900 .byte $info{endprologue}-$info{start_label} 901 .byte $info{num_codes} 902 .byte $frame_encoded 903$info{unwind_codes} 904____ 905 906 # UNWIND_INFOs must be 4-byte aligned. If needed, we must add an extra 907 # unwind code. This does not change the unwind code count. Windows 908 # documentation says "For alignment purposes, this array always has an 909 # even number of entries, and the final entry is potentially unused. In 910 # that case, the array is one longer than indicated by the count of 911 # unwind codes field." 912 if ($info{num_codes} & 1) { 913 $xdata .= "\t.value\t0\n"; 914 } 915 916 %info = (); 917 return $end_label; 918 } 919 920 sub re { 921 my ($class, $line) = @_; 922 if ($$line =~ s/^\s*\.seh_(\w+)\s*//) { 923 my $dir = $1; 924 if (!$win64) { 925 $$line = ""; 926 return; 927 } 928 929 my $label; 930 SWITCH: for ($dir) { 931 /^startproc$/ && do { 932 $label = _startproc($1); 933 last; 934 }; 935 /^pushreg$/ && do { 936 $$line =~ /^(%\w+)\s*$/ or die "could not parse .seh_$dir"; 937 my $reg_num = $UWOP_REG_TO_NUMBER{$1} or die "unknown register $1"; 938 _updating_fixed_allocation(); 939 $label = _add_unwind_code($UWOP_PUSH_NONVOL, $reg_num); 940 last; 941 }; 942 /^stackalloc$/ && do { 943 my $num = eval($$line); 944 if ($num <= 0 || $num % 8 != 0) { 945 die "invalid stack allocation: $num"; 946 } 947 _updating_fixed_allocation(); 948 if ($num <= 128) { 949 $label = _add_unwind_code($UWOP_ALLOC_SMALL, ($num - 8) / 8); 950 } elsif ($num < 512 * 1024) { 951 $label = _add_unwind_code($UWOP_ALLOC_LARGE, 0, $num / 8); 952 } elsif ($num < 4 * 1024 * 1024 * 1024) { 953 $label = _add_unwind_code($UWOP_ALLOC_LARGE, 1, $num >> 16, $num & 0xffff); 954 } else { 955 die "stack allocation too large: $num" 956 } 957 last; 958 }; 959 /^setframe$/ && do { 960 if ($info{frame_reg} != 0) { 961 die "duplicate .seh_setframe directive"; 962 } 963 if ($info{has_offset}) { 964 die "directives with with an offset must come after .seh_setframe."; 965 } 966 $$line =~ /(%\w+)\s*,\s*(.+)/ or die "could not parse .seh_$dir"; 967 my $reg_num = $UWOP_REG_TO_NUMBER{$1} or die "unknown register $1"; 968 my $offset = eval($2); 969 if ($offset < 0 || $offset % 16 != 0 || $offset > 240) { 970 die "invalid offset: $offset"; 971 } 972 $info{frame_reg} = $reg_num; 973 $info{frame_offset} = $offset; 974 $label = _add_unwind_code($UWOP_SET_FPREG, 0); 975 last; 976 }; 977 /^savereg$/ && do { 978 $$line =~ /(%\w+)\s*,\s*(.+)/ or die "could not parse .seh_$dir"; 979 my $reg_num = $UWOP_REG_TO_NUMBER{$1} or die "unknown register $1"; 980 my $offset = eval($2); 981 if ($offset < 0 || $offset % 8 != 0) { 982 die "invalid offset: $offset"; 983 } 984 if ($offset < 8 * 65536) { 985 $label = _add_unwind_code($UWOP_SAVE_NONVOL, $reg_num, $offset / 8); 986 } else { 987 $label = _add_unwind_code($UWOP_SAVE_NONVOL_FAR, $reg_num, $offset >> 16, $offset & 0xffff); 988 } 989 $info{has_offset} = 1; 990 last; 991 }; 992 /^savexmm$/ && do { 993 $$line =~ /%xmm(\d+)\s*,\s*(.+)/ or die "could not parse .seh_$dir"; 994 my $reg_num = $1; 995 my $offset = eval($2); 996 if ($offset < 0 || $offset % 16 != 0) { 997 die "invalid offset: $offset"; 998 } 999 if ($offset < 16 * 65536) { 1000 $label = _add_unwind_code($UWOP_SAVE_XMM128, $reg_num, $offset / 16); 1001 } else { 1002 $label = _add_unwind_code($UWOP_SAVE_XMM128_FAR, $reg_num, $offset >> 16, $offset & 0xffff); 1003 } 1004 $info{has_offset} = 1; 1005 last; 1006 }; 1007 /^endprologue$/ && do { 1008 _check_in_prologue(); 1009 if ($info{num_codes} == 0) { 1010 # If a Windows function has no directives (i.e. it 1011 # doesn't touch the stack), it is a leaf function and is 1012 # not expected to appear in .pdata or .xdata. 1013 die ".seh_endprologue found with no unwind codes"; 1014 } 1015 1016 $label = _new_unwind_label("endprologue"); 1017 $info{endprologue} = $label; 1018 last; 1019 }; 1020 /^endproc$/ && do { 1021 $label = _endproc(); 1022 last; 1023 }; 1024 die "unknown SEH directive .seh_$dir"; 1025 } 1026 1027 # All SEH directives compile to labels inline. The other data is 1028 # emitted later. 1029 $$line = ""; 1030 $label .= ":"; 1031 return label->re(\$label); 1032 } 1033 } 1034 1035 sub pdata_and_xdata { 1036 return "" unless $win64; 1037 1038 my $ret = ""; 1039 if ($pdata ne "") { 1040 $ret .= <<____; 1041.section .pdata 1042.align 4 1043$pdata 1044____ 1045 } 1046 if ($xdata ne "") { 1047 $ret .= <<____; 1048.section .xdata 1049.align 4 1050$xdata 1051____ 1052 } 1053 return $ret; 1054 } 1055} 1056{ package directive; # pick up directives, which start with . 1057 my %sections; 1058 sub nasm_section { 1059 my ($name, $qualifiers) = @_; 1060 my $ret = "section\t$name"; 1061 if (exists $sections{$name}) { 1062 # Work around https://bugzilla.nasm.us/show_bug.cgi?id=3392701. Only 1063 # emit section qualifiers the first time a section is referenced. 1064 # For all subsequent references, require the qualifiers match and 1065 # omit them. 1066 # 1067 # See also https://crbug.com/1422018 and b/270643835. 1068 my $old = $sections{$name}; 1069 die "Inconsistent qualifiers: $qualifiers vs $old" if ($qualifiers ne "" && $qualifiers ne $old); 1070 } else { 1071 $sections{$name} = $qualifiers; 1072 if ($qualifiers ne "") { 1073 $ret .= " $qualifiers"; 1074 } 1075 } 1076 return $ret; 1077 } 1078 sub re { 1079 my ($class, $line) = @_; 1080 my $self = {}; 1081 my $ret; 1082 my $dir; 1083 1084 # chain-call to cfi_directive and seh_directive. 1085 $ret = cfi_directive->re($line) and return $ret; 1086 $ret = seh_directive->re($line) and return $ret; 1087 1088 if ($$line =~ /^\s*(\.\w+)/) { 1089 bless $self,$class; 1090 $dir = $1; 1091 $ret = $self; 1092 undef $self->{value}; 1093 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//; 1094 1095 SWITCH: for ($dir) { 1096 /\.global|\.globl|\.extern/ 1097 && do { $globals{$$line} = $prefix . $$line; 1098 $$line = $globals{$$line} if ($prefix); 1099 last; 1100 }; 1101 /\.type/ && do { my ($sym,$type,$narg) = split(/\s*,\s*/,$$line); 1102 if ($type eq "\@function") { 1103 undef $current_function; 1104 $current_function->{name} = $sym; 1105 $current_function->{abi} = "svr4"; 1106 $current_function->{narg} = $narg; 1107 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 1108 } elsif ($type eq "\@abi-omnipotent") { 1109 undef $current_function; 1110 $current_function->{name} = $sym; 1111 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; 1112 } 1113 $$line =~ s/\@abi\-omnipotent/\@function/; 1114 $$line =~ s/\@function.*/\@function/; 1115 last; 1116 }; 1117 /\.asciz/ && do { if ($$line =~ /^"(.*)"$/) { 1118 $dir = ".byte"; 1119 $$line = join(",",unpack("C*",$1),0); 1120 } 1121 last; 1122 }; 1123 /\.rva|\.long|\.quad|\.byte/ 1124 && do { $$line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; 1125 $$line =~ s/\.L/$decor/g; 1126 last; 1127 }; 1128 } 1129 1130 if ($gas) { 1131 $self->{value} = $dir . "\t" . $$line; 1132 1133 if ($dir =~ /\.extern/) { 1134 if ($flavour eq "elf") { 1135 $self->{value} .= "\n.hidden $$line"; 1136 } else { 1137 $self->{value} = ""; 1138 } 1139 } elsif (!$elf && $dir =~ /\.type/) { 1140 $self->{value} = ""; 1141 $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" . 1142 (defined($globals{$1})?".scl 2;":".scl 3;") . 1143 "\t.type 32;\t.endef" 1144 if ($win64 && $$line =~ /([^,]+),\@function/); 1145 } elsif (!$elf && $dir =~ /\.size/) { 1146 $self->{value} = ""; 1147 if (defined($current_function)) { 1148 $self->{value} .= "${decor}SEH_end_$current_function->{name}:" 1149 if ($win64 && $current_function->{abi} eq "svr4"); 1150 undef $current_function; 1151 } 1152 } elsif (!$elf && $dir =~ /\.align/) { 1153 $self->{value} = ".p2align\t" . (log($$line)/log(2)); 1154 } elsif ($dir eq ".section") { 1155 $current_segment=$$line; 1156 if (!$elf && $current_segment eq ".rodata") { 1157 if ($flavour eq "macosx") { $self->{value} = ".section\t__DATA,__const"; } 1158 } 1159 if (!$elf && $current_segment eq ".init") { 1160 if ($flavour eq "macosx") { $self->{value} = ".mod_init_func"; } 1161 elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.ctors"; } 1162 } 1163 } elsif ($dir =~ /\.(text|data)/) { 1164 $current_segment=".$1"; 1165 } elsif ($dir =~ /\.global|\.globl|\.extern/) { 1166 if ($flavour eq "macosx") { 1167 $self->{value} .= "\n.private_extern $$line"; 1168 } else { 1169 $self->{value} .= "\n.hidden $$line"; 1170 } 1171 } elsif ($dir =~ /\.hidden/) { 1172 if ($flavour eq "macosx") { $self->{value} = ".private_extern\t$prefix$$line"; } 1173 elsif ($flavour eq "mingw64") { $self->{value} = ""; } 1174 } elsif ($dir =~ /\.comm/) { 1175 $self->{value} = "$dir\t$prefix$$line"; 1176 $self->{value} =~ s|,([0-9]+),([0-9]+)$|",$1,".log($2)/log(2)|e if ($flavour eq "macosx"); 1177 } 1178 $$line = ""; 1179 return $self; 1180 } 1181 1182 # non-gas case or nasm/masm 1183 SWITCH: for ($dir) { 1184 /\.text/ && do { my $v=undef; 1185 if ($nasm) { 1186 $v=nasm_section(".text", "code align=64")."\n"; 1187 } else { 1188 $v="$current_segment\tENDS\n" if ($current_segment); 1189 $current_segment = ".text\$"; 1190 $v.="$current_segment\tSEGMENT "; 1191 $v.=$masm>=$masmref ? "ALIGN(256)" : "PAGE"; 1192 $v.=" 'CODE'"; 1193 } 1194 $self->{value} = $v; 1195 last; 1196 }; 1197 /\.data/ && do { my $v=undef; 1198 if ($nasm) { 1199 $v=nasm_section(".data", "data align=8")."\n"; 1200 } else { 1201 $v="$current_segment\tENDS\n" if ($current_segment); 1202 $current_segment = "_DATA"; 1203 $v.="$current_segment\tSEGMENT"; 1204 } 1205 $self->{value} = $v; 1206 last; 1207 }; 1208 /\.section/ && do { my $v=undef; 1209 $$line =~ s/([^,]*).*/$1/; 1210 $$line = ".CRT\$XCU" if ($$line eq ".init"); 1211 $$line = ".rdata" if ($$line eq ".rodata"); 1212 if ($nasm) { 1213 my $qualifiers = ""; 1214 if ($$line=~/\.([prx])data/) { 1215 $qualifiers = "rdata align="; 1216 $qualifiers .= $1 eq "p"? 4 : 8; 1217 } elsif ($$line=~/\.CRT\$/i) { 1218 $qualifiers = "rdata align=8"; 1219 } 1220 $v = nasm_section($$line, $qualifiers); 1221 } else { 1222 $v="$current_segment\tENDS\n" if ($current_segment); 1223 $v.="$$line\tSEGMENT"; 1224 if ($$line=~/\.([prx])data/) { 1225 $v.=" READONLY"; 1226 $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref); 1227 } elsif ($$line=~/\.CRT\$/i) { 1228 $v.=" READONLY "; 1229 $v.=$masm>=$masmref ? "ALIGN(8)" : "DWORD"; 1230 } 1231 } 1232 $current_segment = $$line; 1233 $self->{value} = $v; 1234 last; 1235 }; 1236 /\.extern/ && do { $self->{value} = "EXTERN\t".$$line; 1237 $self->{value} .= ":NEAR" if ($masm); 1238 last; 1239 }; 1240 /\.globl|.global/ 1241 && do { $self->{value} = $masm?"PUBLIC":"global"; 1242 $self->{value} .= "\t".$$line; 1243 last; 1244 }; 1245 /\.size/ && do { if (defined($current_function)) { 1246 undef $self->{value}; 1247 if ($current_function->{abi} eq "svr4") { 1248 $self->{value}="${decor}SEH_end_$current_function->{name}:"; 1249 $self->{value}.=":\n" if($masm); 1250 } 1251 $self->{value}.="$current_function->{name}\tENDP" if($masm && $current_function->{name}); 1252 undef $current_function; 1253 } 1254 last; 1255 }; 1256 /\.align/ && do { my $max = ($masm && $masm>=$masmref) ? 256 : 4096; 1257 $self->{value} = "ALIGN\t".($$line>$max?$max:$$line); 1258 last; 1259 }; 1260 /\.(value|long|rva|quad)/ 1261 && do { my $sz = substr($1,0,1); 1262 my @arr = split(/,\s*/,$$line); 1263 my $last = pop(@arr); 1264 my $conv = sub { my $var=shift; 1265 $var=~s/^(0b[0-1]+)/oct($1)/eig; 1266 $var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm); 1267 if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva")) 1268 { $var=~s/^([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; } 1269 $var; 1270 }; 1271 1272 $sz =~ tr/bvlrq/BWDDQ/; 1273 $self->{value} = "\tD$sz\t"; 1274 for (@arr) { $self->{value} .= &$conv($_).","; } 1275 $self->{value} .= &$conv($last); 1276 last; 1277 }; 1278 /\.byte/ && do { my @str=split(/,\s*/,$$line); 1279 map(s/(0b[0-1]+)/oct($1)/eig,@str); 1280 map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm); 1281 while ($#str>15) { 1282 $self->{value}.="\tDB\t" 1283 .join(",",@str[0..15])."\n"; 1284 foreach (0..15) { shift @str; } 1285 } 1286 $self->{value}.="\tDB\t" 1287 .join(",",@str) if (@str); 1288 last; 1289 }; 1290 /\.comm/ && do { my @str=split(/,\s*/,$$line); 1291 my $v=undef; 1292 if ($nasm) { 1293 $v.="common $prefix@str[0] @str[1]"; 1294 } else { 1295 $v="$current_segment\tENDS\n" if ($current_segment); 1296 $current_segment = "_DATA"; 1297 $v.="$current_segment\tSEGMENT\n"; 1298 $v.="COMM @str[0]:DWORD:".@str[1]/4; 1299 } 1300 $self->{value} = $v; 1301 last; 1302 }; 1303 } 1304 $$line = ""; 1305 } 1306 1307 $ret; 1308 } 1309 sub out { 1310 my $self = shift; 1311 $self->{value}; 1312 } 1313} 1314 1315# Upon initial x86_64 introduction SSE>2 extensions were not introduced 1316# yet. In order not to be bothered by tracing exact assembler versions, 1317# but at the same time to provide a bare security minimum of AES-NI, we 1318# hard-code some instructions. Extensions past AES-NI on the other hand 1319# are traced by examining assembler version in individual perlasm 1320# modules... 1321 1322my %regrm = ( "%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3, 1323 "%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7 ); 1324 1325sub rex { 1326 my $opcode=shift; 1327 my ($dst,$src,$rex)=@_; 1328 1329 $rex|=0x04 if($dst>=8); 1330 $rex|=0x01 if($src>=8); 1331 push @$opcode,($rex|0x40) if ($rex); 1332} 1333 1334my $movq = sub { # elderly gas can't handle inter-register movq 1335 my $arg = shift; 1336 my @opcode=(0x66); 1337 if ($arg =~ /%xmm([0-9]+),\s*%r(\w+)/) { 1338 my ($src,$dst)=($1,$2); 1339 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 1340 rex(\@opcode,$src,$dst,0x8); 1341 push @opcode,0x0f,0x7e; 1342 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 1343 @opcode; 1344 } elsif ($arg =~ /%r(\w+),\s*%xmm([0-9]+)/) { 1345 my ($src,$dst)=($2,$1); 1346 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 1347 rex(\@opcode,$src,$dst,0x8); 1348 push @opcode,0x0f,0x6e; 1349 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 1350 @opcode; 1351 } else { 1352 (); 1353 } 1354}; 1355 1356my $pextrd = sub { 1357 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*(%\w+)/) { 1358 my @opcode=(0x66); 1359 my $imm=$1; 1360 my $src=$2; 1361 my $dst=$3; 1362 if ($dst =~ /%r([0-9]+)d/) { $dst = $1; } 1363 elsif ($dst =~ /%e/) { $dst = $regrm{$dst}; } 1364 rex(\@opcode,$src,$dst); 1365 push @opcode,0x0f,0x3a,0x16; 1366 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M 1367 push @opcode,$imm; 1368 @opcode; 1369 } else { 1370 (); 1371 } 1372}; 1373 1374my $pinsrd = sub { 1375 if (shift =~ /\$([0-9]+),\s*(%\w+),\s*%xmm([0-9]+)/) { 1376 my @opcode=(0x66); 1377 my $imm=$1; 1378 my $src=$2; 1379 my $dst=$3; 1380 if ($src =~ /%r([0-9]+)/) { $src = $1; } 1381 elsif ($src =~ /%e/) { $src = $regrm{$src}; } 1382 rex(\@opcode,$dst,$src); 1383 push @opcode,0x0f,0x3a,0x22; 1384 push @opcode,0xc0|(($dst&7)<<3)|($src&7); # ModR/M 1385 push @opcode,$imm; 1386 @opcode; 1387 } else { 1388 (); 1389 } 1390}; 1391 1392my $pshufb = sub { 1393 if (shift =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) { 1394 my @opcode=(0x66); 1395 rex(\@opcode,$2,$1); 1396 push @opcode,0x0f,0x38,0x00; 1397 push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M 1398 @opcode; 1399 } else { 1400 (); 1401 } 1402}; 1403 1404my $palignr = sub { 1405 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 1406 my @opcode=(0x66); 1407 rex(\@opcode,$3,$2); 1408 push @opcode,0x0f,0x3a,0x0f; 1409 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 1410 push @opcode,$1; 1411 @opcode; 1412 } else { 1413 (); 1414 } 1415}; 1416 1417my $pclmulqdq = sub { 1418 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 1419 my @opcode=(0x66); 1420 rex(\@opcode,$3,$2); 1421 push @opcode,0x0f,0x3a,0x44; 1422 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 1423 my $c=$1; 1424 push @opcode,$c=~/^0/?oct($c):$c; 1425 @opcode; 1426 } else { 1427 (); 1428 } 1429}; 1430 1431my $rdrand = sub { 1432 if (shift =~ /%[er](\w+)/) { 1433 my @opcode=(); 1434 my $dst=$1; 1435 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 1436 rex(\@opcode,0,$dst,8); 1437 push @opcode,0x0f,0xc7,0xf0|($dst&7); 1438 @opcode; 1439 } else { 1440 (); 1441 } 1442}; 1443 1444my $rdseed = sub { 1445 if (shift =~ /%[er](\w+)/) { 1446 my @opcode=(); 1447 my $dst=$1; 1448 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } 1449 rex(\@opcode,0,$dst,8); 1450 push @opcode,0x0f,0xc7,0xf8|($dst&7); 1451 @opcode; 1452 } else { 1453 (); 1454 } 1455}; 1456 1457# Not all AVX-capable assemblers recognize AMD XOP extension. Since we 1458# are using only two instructions hand-code them in order to be excused 1459# from chasing assembler versions... 1460 1461sub rxb { 1462 my $opcode=shift; 1463 my ($dst,$src1,$src2,$rxb)=@_; 1464 1465 $rxb|=0x7<<5; 1466 $rxb&=~(0x04<<5) if($dst>=8); 1467 $rxb&=~(0x01<<5) if($src1>=8); 1468 $rxb&=~(0x02<<5) if($src2>=8); 1469 push @$opcode,$rxb; 1470} 1471 1472my $vprotd = sub { 1473 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 1474 my @opcode=(0x8f); 1475 rxb(\@opcode,$3,$2,-1,0x08); 1476 push @opcode,0x78,0xc2; 1477 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 1478 my $c=$1; 1479 push @opcode,$c=~/^0/?oct($c):$c; 1480 @opcode; 1481 } else { 1482 (); 1483 } 1484}; 1485 1486my $vprotq = sub { 1487 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 1488 my @opcode=(0x8f); 1489 rxb(\@opcode,$3,$2,-1,0x08); 1490 push @opcode,0x78,0xc3; 1491 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 1492 my $c=$1; 1493 push @opcode,$c=~/^0/?oct($c):$c; 1494 @opcode; 1495 } else { 1496 (); 1497 } 1498}; 1499 1500# Intel Control-flow Enforcement Technology extension. All functions and 1501# indirect branch targets will have to start with this instruction... 1502 1503my $endbranch = sub { 1504 (0xf3,0x0f,0x1e,0xfa); 1505}; 1506 1507######################################################################## 1508 1509{ 1510 my $comment = "//"; 1511 $comment = ";" if ($masm || $nasm); 1512 print <<___; 1513$comment This file is generated from a similarly-named Perl script in the BoringSSL 1514$comment source tree. Do not edit by hand. 1515 1516___ 1517} 1518 1519if ($nasm) { 1520 die "unknown target" unless ($win64); 1521 print <<___; 1522\%ifidn __OUTPUT_FORMAT__, win64 1523default rel 1524\%define XMMWORD 1525\%define YMMWORD 1526\%define ZMMWORD 1527\%define _CET_ENDBR 1528 1529\%ifdef BORINGSSL_PREFIX 1530\%include "boringssl_prefix_symbols_nasm.inc" 1531\%endif 1532___ 1533} elsif ($masm) { 1534 print <<___; 1535OPTION DOTNAME 1536___ 1537} 1538 1539if ($gas) { 1540 my $target; 1541 if ($elf) { 1542 # The "elf" target is really ELF with SysV ABI, but every ELF platform 1543 # uses the SysV ABI. 1544 $target = "defined(__ELF__)"; 1545 } elsif ($apple) { 1546 $target = "defined(__APPLE__)"; 1547 } else { 1548 die "unknown target: $flavour"; 1549 } 1550 print <<___; 1551#include <openssl/asm_base.h> 1552 1553#if !defined(OPENSSL_NO_ASM) && defined(OPENSSL_X86_64) && $target 1554___ 1555} 1556 1557sub process_line { 1558 my $line = shift; 1559 $line =~ s|\R$||; # Better chomp 1560 1561 if ($nasm) { 1562 $line =~ s|^#ifdef |%ifdef |; 1563 $line =~ s|^#ifndef |%ifndef |; 1564 $line =~ s|^#endif|%endif|; 1565 $line =~ s|[#!].*$||; # get rid of asm-style comments... 1566 } else { 1567 # Get rid of asm-style comments but not preprocessor directives. The 1568 # former are identified by having a letter after the '#' and starting in 1569 # the first column. 1570 $line =~ s|!.*$||; 1571 $line =~ s|(?<=.)#.*$||; 1572 $line =~ s|^#([^a-z].*)?$||; 1573 } 1574 1575 $line =~ s|/\*.*\*/||; # ... and C-style comments... 1576 $line =~ s|^\s+||; # ... and skip white spaces in beginning 1577 $line =~ s|\s+$||; # ... and at the end 1578 1579 if (my $label=label->re(\$line)) { print $label->out(); } 1580 1581 if (my $directive=directive->re(\$line)) { 1582 printf "%s",$directive->out(); 1583 } elsif (my $opcode=opcode->re(\$line)) { 1584 my $asm = eval("\$".$opcode->mnemonic()); 1585 1586 if ((ref($asm) eq 'CODE') && scalar(my @bytes=&$asm($line))) { 1587 print $gas?".byte\t":"DB\t",join(',',@bytes),"\n"; 1588 next; 1589 } 1590 1591 my @args; 1592 ARGUMENT: while (1) { 1593 my $arg; 1594 1595 ($arg=register->re(\$line, $opcode))|| 1596 ($arg=const->re(\$line)) || 1597 ($arg=ea->re(\$line, $opcode)) || 1598 ($arg=expr->re(\$line, $opcode)) || 1599 last ARGUMENT; 1600 1601 push @args,$arg; 1602 1603 last ARGUMENT if ($line !~ /^,/); 1604 1605 $line =~ s/^,\s*//; 1606 } # ARGUMENT: 1607 1608 if ($#args>=0) { 1609 my $insn; 1610 my $sz=$opcode->size(); 1611 1612 if ($gas) { 1613 $insn = $opcode->out($#args>=1?$args[$#args]->size():$sz); 1614 @args = map($_->out($sz),@args); 1615 printf "\t%s\t%s",$insn,join(",",@args); 1616 } else { 1617 $insn = $opcode->out(); 1618 foreach (@args) { 1619 my $arg = $_->out(); 1620 # $insn.=$sz compensates for movq, pinsrw, ... 1621 if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; } 1622 if ($arg =~ /^ymm[0-9]+$/) { $insn.=$sz; $sz="y" if(!$sz); last; } 1623 if ($arg =~ /^zmm[0-9]+$/) { $insn.=$sz; $sz="z" if(!$sz); last; } 1624 if ($arg =~ /^mm[0-9]+$/) { $insn.=$sz; $sz="q" if(!$sz); last; } 1625 } 1626 @args = reverse(@args); 1627 undef $sz if ($nasm && $opcode->mnemonic() eq "lea"); 1628 printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args)); 1629 } 1630 } else { 1631 printf "\t%s",$opcode->out(); 1632 } 1633 } 1634 1635 print $line,"\n"; 1636} 1637 1638while(defined(my $line=<>)) { 1639 process_line($line); 1640} 1641foreach my $line (split(/\n/, seh_directive->pdata_and_xdata())) { 1642 process_line($line); 1643} 1644 1645print "\n$current_segment\tENDS\n" if ($current_segment && $masm); 1646if ($masm) { 1647 print "END\n"; 1648} elsif ($gas) { 1649 print "#endif\n"; 1650} elsif ($nasm) { 1651 print <<___; 1652\%else 1653; Work around https://bugzilla.nasm.us/show_bug.cgi?id=3392738 1654ret 1655\%endif 1656___ 1657} else { 1658 die "unknown assembler"; 1659} 1660 1661close STDOUT or die "error closing STDOUT: $!"; 1662 1663 ################################################# 1664# Cross-reference x86_64 ABI "card" 1665# 1666# Unix Win64 1667# %rax * * 1668# %rbx - - 1669# %rcx #4 #1 1670# %rdx #3 #2 1671# %rsi #2 - 1672# %rdi #1 - 1673# %rbp - - 1674# %rsp - - 1675# %r8 #5 #3 1676# %r9 #6 #4 1677# %r10 * * 1678# %r11 * * 1679# %r12 - - 1680# %r13 - - 1681# %r14 - - 1682# %r15 - - 1683# 1684# (*) volatile register 1685# (-) preserved by callee 1686# (#) Nth argument, volatile 1687# 1688# In Unix terms top of stack is argument transfer area for arguments 1689# which could not be accommodated in registers. Or in other words 7th 1690# [integer] argument resides at 8(%rsp) upon function entry point. 1691# 128 bytes above %rsp constitute a "red zone" which is not touched 1692# by signal handlers and can be used as temporal storage without 1693# allocating a frame. 1694# 1695# In Win64 terms N*8 bytes on top of stack is argument transfer area, 1696# which belongs to/can be overwritten by callee. N is the number of 1697# arguments passed to callee, *but* not less than 4! This means that 1698# upon function entry point 5th argument resides at 40(%rsp), as well 1699# as that 32 bytes from 8(%rsp) can always be used as temporal 1700# storage [without allocating a frame]. One can actually argue that 1701# one can assume a "red zone" above stack pointer under Win64 as well. 1702# Point is that at apparently no occasion Windows kernel would alter 1703# the area above user stack pointer in true asynchronous manner... 1704# 1705# All the above means that if assembler programmer adheres to Unix 1706# register and stack layout, but disregards the "red zone" existence, 1707# it's possible to use following prologue and epilogue to "gear" from 1708# Unix to Win64 ABI in leaf functions with not more than 6 arguments. 1709# 1710# omnipotent_function: 1711# ifdef WIN64 1712# movq %rdi,8(%rsp) 1713# movq %rsi,16(%rsp) 1714# movq %rcx,%rdi ; if 1st argument is actually present 1715# movq %rdx,%rsi ; if 2nd argument is actually ... 1716# movq %r8,%rdx ; if 3rd argument is ... 1717# movq %r9,%rcx ; if 4th argument ... 1718# movq 40(%rsp),%r8 ; if 5th ... 1719# movq 48(%rsp),%r9 ; if 6th ... 1720# endif 1721# ... 1722# ifdef WIN64 1723# movq 8(%rsp),%rdi 1724# movq 16(%rsp),%rsi 1725# endif 1726# ret 1727# 1728 ################################################# 1729# Win64 SEH, Structured Exception Handling. 1730# 1731# Unlike on Unix systems(*) lack of Win64 stack unwinding information 1732# has undesired side-effect at run-time: if an exception is raised in 1733# assembler subroutine such as those in question (basically we're 1734# referring to segmentation violations caused by malformed input 1735# parameters), the application is briskly terminated without invoking 1736# any exception handlers, most notably without generating memory dump 1737# or any user notification whatsoever. This poses a problem. It's 1738# possible to address it by registering custom language-specific 1739# handler that would restore processor context to the state at 1740# subroutine entry point and return "exception is not handled, keep 1741# unwinding" code. Writing such handler can be a challenge... But it's 1742# doable, though requires certain coding convention. Consider following 1743# snippet: 1744# 1745# .type function,@function 1746# function: 1747# movq %rsp,%rax # copy rsp to volatile register 1748# pushq %r15 # save non-volatile registers 1749# pushq %rbx 1750# pushq %rbp 1751# movq %rsp,%r11 1752# subq %rdi,%r11 # prepare [variable] stack frame 1753# andq $-64,%r11 1754# movq %rax,0(%r11) # check for exceptions 1755# movq %r11,%rsp # allocate [variable] stack frame 1756# movq %rax,0(%rsp) # save original rsp value 1757# magic_point: 1758# ... 1759# movq 0(%rsp),%rcx # pull original rsp value 1760# movq -24(%rcx),%rbp # restore non-volatile registers 1761# movq -16(%rcx),%rbx 1762# movq -8(%rcx),%r15 1763# movq %rcx,%rsp # restore original rsp 1764# magic_epilogue: 1765# ret 1766# .size function,.-function 1767# 1768# The key is that up to magic_point copy of original rsp value remains 1769# in chosen volatile register and no non-volatile register, except for 1770# rsp, is modified. While past magic_point rsp remains constant till 1771# the very end of the function. In this case custom language-specific 1772# exception handler would look like this: 1773# 1774# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, 1775# CONTEXT *context,DISPATCHER_CONTEXT *disp) 1776# { ULONG64 *rsp = (ULONG64 *)context->Rax; 1777# ULONG64 rip = context->Rip; 1778# 1779# if (rip >= magic_point) 1780# { rsp = (ULONG64 *)context->Rsp; 1781# if (rip < magic_epilogue) 1782# { rsp = (ULONG64 *)rsp[0]; 1783# context->Rbp = rsp[-3]; 1784# context->Rbx = rsp[-2]; 1785# context->R15 = rsp[-1]; 1786# } 1787# } 1788# context->Rsp = (ULONG64)rsp; 1789# context->Rdi = rsp[1]; 1790# context->Rsi = rsp[2]; 1791# 1792# memcpy (disp->ContextRecord,context,sizeof(CONTEXT)); 1793# RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase, 1794# dips->ControlPc,disp->FunctionEntry,disp->ContextRecord, 1795# &disp->HandlerData,&disp->EstablisherFrame,NULL); 1796# return ExceptionContinueSearch; 1797# } 1798# 1799# It's appropriate to implement this handler in assembler, directly in 1800# function's module. In order to do that one has to know members' 1801# offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant 1802# values. Here they are: 1803# 1804# CONTEXT.Rax 120 1805# CONTEXT.Rcx 128 1806# CONTEXT.Rdx 136 1807# CONTEXT.Rbx 144 1808# CONTEXT.Rsp 152 1809# CONTEXT.Rbp 160 1810# CONTEXT.Rsi 168 1811# CONTEXT.Rdi 176 1812# CONTEXT.R8 184 1813# CONTEXT.R9 192 1814# CONTEXT.R10 200 1815# CONTEXT.R11 208 1816# CONTEXT.R12 216 1817# CONTEXT.R13 224 1818# CONTEXT.R14 232 1819# CONTEXT.R15 240 1820# CONTEXT.Rip 248 1821# CONTEXT.Xmm6 512 1822# sizeof(CONTEXT) 1232 1823# DISPATCHER_CONTEXT.ControlPc 0 1824# DISPATCHER_CONTEXT.ImageBase 8 1825# DISPATCHER_CONTEXT.FunctionEntry 16 1826# DISPATCHER_CONTEXT.EstablisherFrame 24 1827# DISPATCHER_CONTEXT.TargetIp 32 1828# DISPATCHER_CONTEXT.ContextRecord 40 1829# DISPATCHER_CONTEXT.LanguageHandler 48 1830# DISPATCHER_CONTEXT.HandlerData 56 1831# UNW_FLAG_NHANDLER 0 1832# ExceptionContinueSearch 1 1833# 1834# In order to tie the handler to the function one has to compose 1835# couple of structures: one for .xdata segment and one for .pdata. 1836# 1837# UNWIND_INFO structure for .xdata segment would be 1838# 1839# function_unwind_info: 1840# .byte 9,0,0,0 1841# .rva handler 1842# 1843# This structure designates exception handler for a function with 1844# zero-length prologue, no stack frame or frame register. 1845# 1846# To facilitate composing of .pdata structures, auto-generated "gear" 1847# prologue copies rsp value to rax and denotes next instruction with 1848# .LSEH_begin_{function_name} label. This essentially defines the SEH 1849# styling rule mentioned in the beginning. Position of this label is 1850# chosen in such manner that possible exceptions raised in the "gear" 1851# prologue would be accounted to caller and unwound from latter's frame. 1852# End of function is marked with respective .LSEH_end_{function_name} 1853# label. To summarize, .pdata segment would contain 1854# 1855# .rva .LSEH_begin_function 1856# .rva .LSEH_end_function 1857# .rva function_unwind_info 1858# 1859# Reference to function_unwind_info from .xdata segment is the anchor. 1860# In case you wonder why references are 32-bit .rvas and not 64-bit 1861# .quads. References put into these two segments are required to be 1862# *relative* to the base address of the current binary module, a.k.a. 1863# image base. No Win64 module, be it .exe or .dll, can be larger than 1864# 2GB and thus such relative references can be and are accommodated in 1865# 32 bits. 1866# 1867# Having reviewed the example function code, one can argue that "movq 1868# %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix 1869# rax would contain an undefined value. If this "offends" you, use 1870# another register and refrain from modifying rax till magic_point is 1871# reached, i.e. as if it was a non-volatile register. If more registers 1872# are required prior [variable] frame setup is completed, note that 1873# nobody says that you can have only one "magic point." You can 1874# "liberate" non-volatile registers by denoting last stack off-load 1875# instruction and reflecting it in finer grade unwind logic in handler. 1876# After all, isn't it why it's called *language-specific* handler... 1877# 1878# SE handlers are also involved in unwinding stack when executable is 1879# profiled or debugged. Profiling implies additional limitations that 1880# are too subtle to discuss here. For now it's sufficient to say that 1881# in order to simplify handlers one should either a) offload original 1882# %rsp to stack (like discussed above); or b) if you have a register to 1883# spare for frame pointer, choose volatile one. 1884# 1885# (*) Note that we're talking about run-time, not debug-time. Lack of 1886# unwind information makes debugging hard on both Windows and 1887# Unix. "Unlike" refers to the fact that on Unix signal handler 1888# will always be invoked, core dumped and appropriate exit code 1889# returned to parent (for user notification). 1890