kris
/
boringssl
1
0
Форк 0
Код Проблеми 0 Запити на злиття 0 Релізи 0 Вікі Активність
Ви не можете вибрати більше 25 тем Теми мають розпочинатися з літери або цифри, можуть містити дефіси (-) і не повинні перевищувати 35 символів.
3384 Коміти
12 Гілки
38 MiB
 
 
 
 
 
 
Дерево: 4b0d0e4c5e
Гілки Теги
${ item.name }
Створити гілку ${ searchTerm }
з '4b0d0e4c5e'
${ noResults }
boringssl/crypto/perlasm/x86_64-xlate.pl

1194 рядки
36 KiB
Неформатований Звинувачення Історія 

  1. #! /usr/bin/env perl

  2. # Copyright 2005-2016 The OpenSSL Project Authors. All Rights Reserved.

  3. #

  4. # Licensed under the OpenSSL license (the "License").  You may not use

  5. # this file except in compliance with the License.  You can obtain a copy

  6. # in the file LICENSE in the source distribution or at

  7. # https://www.openssl.org/source/license.html

  8. 

  9. 

  10. # Ascetic x86_64 AT&T to MASM/NASM assembler translator by <appro>.

  11. #

  12. # Why AT&T to MASM and not vice versa? Several reasons. Because AT&T

  13. # format is way easier to parse. Because it's simpler to "gear" from

  14. # Unix ABI to Windows one [see cross-reference "card" at the end of

  15. # file]. Because Linux targets were available first...

  16. #

  17. # In addition the script also "distills" code suitable for GNU

  18. # assembler, so that it can be compiled with more rigid assemblers,

  19. # such as Solaris /usr/ccs/bin/as.

  20. #

  21. # This translator is not designed to convert *arbitrary* assembler

  22. # code from AT&T format to MASM one. It's designed to convert just

  23. # enough to provide for dual-ABI OpenSSL modules development...

  24. # There *are* limitations and you might have to modify your assembler

  25. # code or this script to achieve the desired result...

  26. #

  27. # Currently recognized limitations:

  28. #

  29. # - can't use multiple ops per line;

  30. #

  31. # Dual-ABI styling rules.

  32. #

  33. # 1. Adhere to Unix register and stack layout [see cross-reference

  34. #    ABI "card" at the end for explanation].

  35. # 2. Forget about "red zone," stick to more traditional blended

  36. #    stack frame allocation. If volatile storage is actually required

  37. #    that is. If not, just leave the stack as is.

  38. # 3. Functions tagged with ".type name,@function" get crafted with

  39. #    unified Win64 prologue and epilogue automatically. If you want

  40. #    to take care of ABI differences yourself, tag functions as

  41. #    ".type name,@abi-omnipotent" instead.

  42. # 4. To optimize the Win64 prologue you can specify number of input

  43. #    arguments as ".type name,@function,N." Keep in mind that if N is

  44. #    larger than 6, then you *have to* write "abi-omnipotent" code,

  45. #    because >6 cases can't be addressed with unified prologue.

  46. # 5. Name local labels as .L*, do *not* use dynamic labels such as 1:

  47. #    (sorry about latter).

  48. # 6. Don't use [or hand-code with .byte] "rep ret." "ret" mnemonic is

  49. #    required to identify the spots, where to inject Win64 epilogue!

  50. #    But on the pros, it's then prefixed with rep automatically:-)

  51. # 7. Stick to explicit ip-relative addressing. If you have to use

  52. #    GOTPCREL addressing, stick to mov symbol@GOTPCREL(%rip),%r??.

  53. #    Both are recognized and translated to proper Win64 addressing

  54. #    modes. To support legacy code a synthetic directive, .picmeup,

  55. #    is implemented. It puts address of the *next* instruction into

  56. #    target register, e.g.:

  57. #

  58. #		.picmeup	%rax

  59. #		lea		.Label-.(%rax),%rax

  60. #

  61. # 8. In order to provide for structured exception handling unified

  62. #    Win64 prologue copies %rsp value to %rax. For further details

  63. #    see SEH paragraph at the end.

  64. # 9. .init segment is allowed to contain calls to functions only.

  65. # a. If function accepts more than 4 arguments *and* >4th argument

  66. #    is declared as non 64-bit value, do clear its upper part.

  67. 

  68. 

  69. use strict;

  70. 

  71. my $flavour = shift;

  72. my $output  = shift;

  73. if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }

  74. 

  75. open STDOUT,">$output" || die "can't open $output: $!"

  76. 	if (defined($output));

  77. 

  78. my $gas=1;	$gas=0 if ($output =~ /\.asm$/);

  79. my $elf=1;	$elf=0 if (!$gas);

  80. my $win64=0;

  81. my $prefix="";

  82. my $decor=".L";

  83. 

  84. my $masmref=8 + 50727*2**-32;	# 8.00.50727 shipped with VS2005

  85. my $masm=0;

  86. my $PTR=" PTR";

  87. 

  88. my $nasmref=2.03;

  89. my $nasm=0;

  90. 

  91. if    ($flavour eq "mingw64")	{ $gas=1; $elf=0; $win64=1;

  92. 				  # TODO(davidben): Before supporting the

  93. 				  # mingw64 perlasm flavour, do away with this

  94. 				  # environment variable check.

  95.                                   die "mingw64 not supported";

  96. 				  $prefix=`echo __USER_LABEL_PREFIX__ | $ENV{CC} -E -P -`;

  97. 				  $prefix =~ s|\R$||; # Better chomp

  98. 				}

  99. elsif ($flavour eq "macosx")	{ $gas=1; $elf=0; $prefix="_"; $decor="L\$"; }

  100. elsif ($flavour eq "masm")	{ $gas=0; $elf=0; $masm=$masmref; $win64=1; $decor="\$L\$"; }

  101. elsif ($flavour eq "nasm")	{ $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $decor="\$L\$"; $PTR=""; }

  102. elsif (!$gas)			{ die "unknown flavour $flavour"; }

  103. 

  104. my $current_segment;

  105. my $current_function;

  106. my %globals;

  107. 

  108. { package opcode;	# pick up opcodes

  109.     sub re {

  110. 	my	($class, $line) = @_;

  111. 	my	$self = {};

  112. 	my	$ret;

  113. 

  114. 	if ($$line =~ /^([a-z][a-z0-9]*)/i) {

  115. 	    bless $self,$class;

  116. 	    $self->{op} = $1;

  117. 	    $ret = $self;

  118. 	    $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;

  119. 

  120. 	    undef $self->{sz};

  121. 	    if ($self->{op} =~ /^(movz)x?([bw]).*/) {	# movz is pain...

  122. 		$self->{op} = $1;

  123. 		$self->{sz} = $2;

  124. 	    } elsif ($self->{op} =~ /call|jmp/) {

  125. 		$self->{sz} = "";

  126. 	    } elsif ($self->{op} =~ /^p/ && $' !~ /^(ush|op|insrw)/) { # SSEn

  127. 		$self->{sz} = "";

  128. 	    } elsif ($self->{op} =~ /^v/) { # VEX

  129. 		$self->{sz} = "";

  130. 	    } elsif ($self->{op} =~ /mov[dq]/ && $$line =~ /%xmm/) {

  131. 		$self->{sz} = "";

  132. 	    } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) {

  133. 		$self->{op} = $1;

  134. 		$self->{sz} = $2;

  135. 	    }

  136. 	}

  137. 	$ret;

  138.     }

  139.     sub size {

  140. 	my ($self, $sz) = @_;

  141. 	$self->{sz} = $sz if (defined($sz) && !defined($self->{sz}));

  142. 	$self->{sz};

  143.     }

  144.     sub out {

  145. 	my $self = shift;

  146. 	if ($gas) {

  147. 	    if ($self->{op} eq "movz") {	# movz is pain...

  148. 		sprintf "%s%s%s",$self->{op},$self->{sz},shift;

  149. 	    } elsif ($self->{op} =~ /^set/) { 

  150. 		"$self->{op}";

  151. 	    } elsif ($self->{op} eq "ret") {

  152. 		my $epilogue = "";

  153. 		if ($win64 && $current_function->{abi} eq "svr4") {

  154. 		    $epilogue = "movq	8(%rsp),%rdi\n\t" .

  155. 				"movq	16(%rsp),%rsi\n\t";

  156. 		}

  157. 	    	$epilogue . ".byte	0xf3,0xc3";

  158. 	    } elsif ($self->{op} eq "call" && !$elf && $current_segment eq ".init") {

  159. 		".p2align\t3\n\t.quad";

  160. 	    } else {

  161. 		"$self->{op}$self->{sz}";

  162. 	    }

  163. 	} else {

  164. 	    $self->{op} =~ s/^movz/movzx/;

  165. 	    if ($self->{op} eq "ret") {

  166. 		$self->{op} = "";

  167. 		if ($win64 && $current_function->{abi} eq "svr4") {

  168. 		    $self->{op} = "mov	rdi,QWORD$PTR\[8+rsp\]\t;WIN64 epilogue\n\t".

  169. 				  "mov	rsi,QWORD$PTR\[16+rsp\]\n\t";

  170. 	    	}

  171. 		$self->{op} .= "DB\t0F3h,0C3h\t\t;repret";

  172. 	    } elsif ($self->{op} =~ /^(pop|push)f/) {

  173. 		$self->{op} .= $self->{sz};

  174. 	    } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") {

  175. 		$self->{op} = "\tDQ";

  176. 	    } 

  177. 	    $self->{op};

  178. 	}

  179.     }

  180.     sub mnemonic {

  181. 	my ($self, $op) = @_;

  182. 	$self->{op}=$op if (defined($op));

  183. 	$self->{op};

  184.     }

  185. }

  186. { package const;	# pick up constants, which start with $

  187.     sub re {

  188. 	my	($class, $line) = @_;

  189. 	my	$self = {};

  190. 	my	$ret;

  191. 

  192. 	if ($$line =~ /^\$([^,]+)/) {

  193. 	    bless $self, $class;

  194. 	    $self->{value} = $1;

  195. 	    $ret = $self;

  196. 	    $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;

  197. 	}

  198. 	$ret;

  199.     }

  200.     sub out {

  201.     	my $self = shift;

  202. 

  203. 	$self->{value} =~ s/\b(0b[0-1]+)/oct($1)/eig;

  204. 	if ($gas) {

  205. 	    # Solaris /usr/ccs/bin/as can't handle multiplications

  206. 	    # in $self->{value}

  207. 	    my $value = $self->{value};

  208. 	    no warnings;    # oct might complain about overflow, ignore here...

  209. 	    $value =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;

  210. 	    if ($value =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg) {

  211. 		$self->{value} = $value;

  212. 	    }

  213. 	    sprintf "\$%s",$self->{value};

  214. 	} else {

  215. 	    $self->{value} =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm);

  216. 	    sprintf "%s",$self->{value};

  217. 	}

  218.     }

  219. }

  220. { package ea;		# pick up effective addresses: expr(%reg,%reg,scale)

  221.     sub re {

  222. 	my	($class, $line, $opcode) = @_;

  223. 	my	$self = {};

  224. 	my	$ret;

  225. 

  226. 	# optional * ----vvv--- appears in indirect jmp/call

  227. 	if ($$line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)/) {

  228. 	    bless $self, $class;

  229. 	    $self->{asterisk} = $1;

  230. 	    $self->{label} = $2;

  231. 	    ($self->{base},$self->{index},$self->{scale})=split(/,/,$3);

  232. 	    $self->{scale} = 1 if (!defined($self->{scale}));

  233. 	    $ret = $self;

  234. 	    $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;

  235. 

  236. 	    if ($win64 && $self->{label} =~ s/\@GOTPCREL//) {

  237. 		die if ($opcode->mnemonic() ne "mov");

  238. 		$opcode->mnemonic("lea");

  239. 	    }

  240. 	    $self->{base}  =~ s/^%//;

  241. 	    $self->{index} =~ s/^%// if (defined($self->{index}));

  242. 	    $self->{opcode} = $opcode;

  243. 	}

  244. 	$ret;

  245.     }

  246.     sub size {}

  247.     sub out {

  248. 	my ($self, $sz) = @_;

  249. 

  250. 	$self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;

  251. 	$self->{label} =~ s/\.L/$decor/g;

  252. 

  253. 	# Silently convert all EAs to 64-bit. This is required for

  254. 	# elder GNU assembler and results in more compact code,

  255. 	# *but* most importantly AES module depends on this feature!

  256. 	$self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;

  257. 	$self->{base}  =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;

  258. 

  259. 	# Solaris /usr/ccs/bin/as can't handle multiplications

  260. 	# in $self->{label}, new gas requires sign extension...

  261. 	use integer;

  262. 	$self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;

  263. 	$self->{label} =~ s/\b([0-9]+\s*[\*\/\%]\s*[0-9]+)\b/eval($1)/eg;

  264. 	$self->{label} =~ s/\b([0-9]+)\b/$1<<32>>32/eg;

  265. 

  266. 	if (!$self->{label} && $self->{index} && $self->{scale}==1 &&

  267. 	    $self->{base} =~ /(rbp|r13)/) {

  268. 		$self->{base} = $self->{index}; $self->{index} = $1;

  269. 	}

  270. 

  271. 	if ($gas) {

  272. 	    $self->{label} =~ s/^___imp_/__imp__/   if ($flavour eq "mingw64");

  273. 

  274. 	    if (defined($self->{index})) {

  275. 		sprintf "%s%s(%s,%%%s,%d)",$self->{asterisk},

  276. 					$self->{label},

  277. 					$self->{base}?"%$self->{base}":"",

  278. 					$self->{index},$self->{scale};

  279. 	    } else {

  280. 		sprintf "%s%s(%%%s)",	$self->{asterisk},$self->{label},$self->{base};

  281. 	    }

  282. 	} else {

  283. 	    my %szmap = (	b=>"BYTE$PTR",  w=>"WORD$PTR",

  284. 			l=>"DWORD$PTR", d=>"DWORD$PTR",

  285. 	    		q=>"QWORD$PTR", o=>"OWORD$PTR",

  286. 			x=>"XMMWORD$PTR", y=>"YMMWORD$PTR", z=>"ZMMWORD$PTR" );

  287. 

  288. 	    $self->{label} =~ s/\./\$/g;

  289. 	    $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig;

  290. 	    $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/);

  291. 

  292. 	    my $mnemonic = $self->{opcode}->mnemonic();

  293. 	    ($self->{asterisk})				&& ($sz="q") ||

  294. 	    ($mnemonic =~ /^v?mov([qd])$/)		&& ($sz=$1)  ||

  295. 	    ($mnemonic =~ /^v?pinsr([qdwb])$/)		&& ($sz=$1)  ||

  296. 	    ($mnemonic =~ /^vpbroadcast([qdwb])$/)	&& ($sz=$1)  ||

  297. 	    ($mnemonic =~ /^v(?!perm)[a-z]+[fi]128$/)	&& ($sz="x");

  298. 

  299. 	    if (defined($self->{index})) {

  300. 		sprintf "%s[%s%s*%d%s]",$szmap{$sz},

  301. 					$self->{label}?"$self->{label}+":"",

  302. 					$self->{index},$self->{scale},

  303. 					$self->{base}?"+$self->{base}":"";

  304. 	    } elsif ($self->{base} eq "rip") {

  305. 		sprintf "%s[%s]",$szmap{$sz},$self->{label};

  306. 	    } else {

  307. 		sprintf "%s[%s%s]",$szmap{$sz},

  308. 					$self->{label}?"$self->{label}+":"",

  309. 					$self->{base};

  310. 	    }

  311. 	}

  312.     }

  313. }

  314. { package register;	# pick up registers, which start with %.

  315.     sub re {

  316. 	my	($class, $line, $opcode) = @_;

  317. 	my	$self = {};

  318. 	my	$ret;

  319. 

  320. 	# optional * ----vvv--- appears in indirect jmp/call

  321. 	if ($$line =~ /^(\*?)%(\w+)/) {

  322. 	    bless $self,$class;

  323. 	    $self->{asterisk} = $1;

  324. 	    $self->{value} = $2;

  325. 	    $opcode->size($self->size());

  326. 	    $ret = $self;

  327. 	    $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;

  328. 	}

  329. 	$ret;

  330.     }

  331.     sub size {

  332. 	my	$self = shift;

  333. 	my	$ret;

  334. 

  335. 	if    ($self->{value} =~ /^r[\d]+b$/i)	{ $ret="b"; }

  336. 	elsif ($self->{value} =~ /^r[\d]+w$/i)	{ $ret="w"; }

  337. 	elsif ($self->{value} =~ /^r[\d]+d$/i)	{ $ret="l"; }

  338. 	elsif ($self->{value} =~ /^r[\w]+$/i)	{ $ret="q"; }

  339. 	elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; }

  340. 	elsif ($self->{value} =~ /^[\w]{2}l$/i)	{ $ret="b"; }

  341. 	elsif ($self->{value} =~ /^[\w]{2}$/i)	{ $ret="w"; }

  342. 	elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; }

  343. 

  344. 	$ret;

  345.     }

  346.     sub out {

  347.     	my $self = shift;

  348. 	if ($gas)	{ sprintf "%s%%%s",$self->{asterisk},$self->{value}; }

  349. 	else		{ $self->{value}; }

  350.     }

  351. }

  352. { package label;	# pick up labels, which end with :

  353.     sub re {

  354. 	my	($class, $line) = @_;

  355. 	my	$self = {};

  356. 	my	$ret;

  357. 

  358. 	if ($$line =~ /(^[\.\w]+)\:/) {

  359. 	    bless $self,$class;

  360. 	    $self->{value} = $1;

  361. 	    $ret = $self;

  362. 	    $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;

  363. 

  364. 	    $self->{value} =~ s/^\.L/$decor/;

  365. 	}

  366. 	$ret;

  367.     }

  368.     sub out {

  369. 	my $self = shift;

  370. 

  371. 	if ($gas) {

  372. 	    my $func = ($globals{$self->{value}} or $self->{value}) . ":";

  373. 	    if ($win64	&&

  374. 			$current_function->{name} eq $self->{value} &&

  375. 			$current_function->{abi} eq "svr4") {

  376. 		$func .= "\n";

  377. 		$func .= "	movq	%rdi,8(%rsp)\n";

  378. 		$func .= "	movq	%rsi,16(%rsp)\n";

  379. 		$func .= "	movq	%rsp,%rax\n";

  380. 		$func .= "${decor}SEH_begin_$current_function->{name}:\n";

  381. 		my $narg = $current_function->{narg};

  382. 		$narg=6 if (!defined($narg));

  383. 		$func .= "	movq	%rcx,%rdi\n" if ($narg>0);

  384. 		$func .= "	movq	%rdx,%rsi\n" if ($narg>1);

  385. 		$func .= "	movq	%r8,%rdx\n"  if ($narg>2);

  386. 		$func .= "	movq	%r9,%rcx\n"  if ($narg>3);

  387. 		$func .= "	movq	40(%rsp),%r8\n" if ($narg>4);

  388. 		$func .= "	movq	48(%rsp),%r9\n" if ($narg>5);

  389. 	    }

  390. 	    $func;

  391. 	} elsif ($self->{value} ne "$current_function->{name}") {

  392. 	    # Make all labels in masm global.

  393. 	    $self->{value} .= ":" if ($masm);

  394. 	    $self->{value} . ":";

  395. 	} elsif ($win64 && $current_function->{abi} eq "svr4") {

  396. 	    my $func =	"$current_function->{name}" .

  397. 			($nasm ? ":" : "\tPROC $current_function->{scope}") .

  398. 			"\n";

  399. 	    $func .= "	mov	QWORD$PTR\[8+rsp\],rdi\t;WIN64 prologue\n";

  400. 	    $func .= "	mov	QWORD$PTR\[16+rsp\],rsi\n";

  401. 	    $func .= "	mov	rax,rsp\n";

  402. 	    $func .= "${decor}SEH_begin_$current_function->{name}:";

  403. 	    $func .= ":" if ($masm);

  404. 	    $func .= "\n";

  405. 	    my $narg = $current_function->{narg};

  406. 	    $narg=6 if (!defined($narg));

  407. 	    $func .= "	mov	rdi,rcx\n" if ($narg>0);

  408. 	    $func .= "	mov	rsi,rdx\n" if ($narg>1);

  409. 	    $func .= "	mov	rdx,r8\n"  if ($narg>2);

  410. 	    $func .= "	mov	rcx,r9\n"  if ($narg>3);

  411. 	    $func .= "	mov	r8,QWORD$PTR\[40+rsp\]\n" if ($narg>4);

  412. 	    $func .= "	mov	r9,QWORD$PTR\[48+rsp\]\n" if ($narg>5);

  413. 	    $func .= "\n";

  414. 	} else {

  415. 	   "$current_function->{name}".

  416. 			($nasm ? ":" : "\tPROC $current_function->{scope}");

  417. 	}

  418.     }

  419. }

  420. { package expr;		# pick up expressioins

  421.     sub re {

  422. 	my	($class, $line, $opcode) = @_;

  423. 	my	$self = {};

  424. 	my	$ret;

  425. 

  426. 	if ($$line =~ /(^[^,]+)/) {

  427. 	    bless $self,$class;

  428. 	    $self->{value} = $1;

  429. 	    $ret = $self;

  430. 	    $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;

  431. 

  432. 	    $self->{value} =~ s/\@PLT// if (!$elf);

  433. 	    $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;

  434. 	    $self->{value} =~ s/\.L/$decor/g;

  435. 	    $self->{opcode} = $opcode;

  436. 	}

  437. 	$ret;

  438.     }

  439.     sub out {

  440. 	my $self = shift;

  441. 	if ($nasm && $self->{opcode}->mnemonic()=~m/^j(?![re]cxz)/) {

  442. 	    "NEAR ".$self->{value};

  443. 	} else {

  444. 	    $self->{value};

  445. 	}

  446.     }

  447. }

  448. { package directive;	# pick up directives, which start with .

  449.     sub re {

  450. 	my	($class, $line) = @_;

  451. 	my	$self = {};

  452. 	my	$ret;

  453. 	my	$dir;

  454. 	my	%opcode =	# lea 2f-1f(%rip),%dst; 1: nop; 2:

  455. 		(	"%rax"=>0x01058d48,	"%rcx"=>0x010d8d48,

  456. 			"%rdx"=>0x01158d48,	"%rbx"=>0x011d8d48,

  457. 			"%rsp"=>0x01258d48,	"%rbp"=>0x012d8d48,

  458. 			"%rsi"=>0x01358d48,	"%rdi"=>0x013d8d48,

  459. 			"%r8" =>0x01058d4c,	"%r9" =>0x010d8d4c,

  460. 			"%r10"=>0x01158d4c,	"%r11"=>0x011d8d4c,

  461. 			"%r12"=>0x01258d4c,	"%r13"=>0x012d8d4c,

  462. 			"%r14"=>0x01358d4c,	"%r15"=>0x013d8d4c	);

  463. 

  464. 	if ($$line =~ /^\s*(\.\w+)/) {

  465. 	    bless $self,$class;

  466. 	    $dir = $1;

  467. 	    $ret = $self;

  468. 	    undef $self->{value};

  469. 	    $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;

  470. 

  471. 	    SWITCH: for ($dir) {

  472. 		/\.picmeup/ && do { if ($$line =~ /(%r[\w]+)/i) {

  473. 			    		$dir="\t.long";

  474. 					$$line=sprintf "0x%x,0x90000000",$opcode{$1};

  475. 				    }

  476. 				    last;

  477. 				  };

  478. 		/\.global|\.globl|\.extern/

  479. 			    && do { $globals{$$line} = $prefix . $$line;

  480. 				    $$line = $globals{$$line} if ($prefix);

  481. 				    last;

  482. 				  };

  483. 		/\.type/    && do { my ($sym,$type,$narg) = split(',',$$line);

  484. 				    if ($type eq "\@function") {

  485. 					undef $current_function;

  486. 					$current_function->{name} = $sym;

  487. 					$current_function->{abi}  = "svr4";

  488. 					$current_function->{narg} = $narg;

  489. 					$current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";

  490. 				    } elsif ($type eq "\@abi-omnipotent") {

  491. 					undef $current_function;

  492. 					$current_function->{name} = $sym;

  493. 					$current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";

  494. 				    }

  495. 				    $$line =~ s/\@abi\-omnipotent/\@function/;

  496. 				    $$line =~ s/\@function.*/\@function/;

  497. 				    last;

  498. 				  };

  499. 		/\.asciz/   && do { if ($$line =~ /^"(.*)"$/) {

  500. 					$dir  = ".byte";

  501. 					$$line = join(",",unpack("C*",$1),0);

  502. 				    }

  503. 				    last;

  504. 				  };

  505. 		/\.rva|\.long|\.quad/

  506. 			    && do { $$line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;

  507. 				    $$line =~ s/\.L/$decor/g;

  508. 				    last;

  509. 				  };

  510. 	    }

  511. 

  512. 	    if ($gas) {

  513. 		$self->{value} = $dir . "\t" . $$line;

  514. 

  515. 		if ($dir =~ /\.extern/) {

  516. 		    if ($flavour eq "elf") {

  517. 			$self->{value} .= "\n.hidden $$line";

  518. 		    } else {

  519. 			$self->{value} = "";

  520. 		    }

  521. 		} elsif (!$elf && $dir =~ /\.type/) {

  522. 		    $self->{value} = "";

  523. 		    $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" .

  524. 				(defined($globals{$1})?".scl 2;":".scl 3;") .

  525. 				"\t.type 32;\t.endef"

  526. 				if ($win64 && $$line =~ /([^,]+),\@function/);

  527. 		} elsif (!$elf && $dir =~ /\.size/) {

  528. 		    $self->{value} = "";

  529. 		    if (defined($current_function)) {

  530. 			$self->{value} .= "${decor}SEH_end_$current_function->{name}:"

  531. 				if ($win64 && $current_function->{abi} eq "svr4");

  532. 			undef $current_function;

  533. 		    }

  534. 		} elsif (!$elf && $dir =~ /\.align/) {

  535. 		    $self->{value} = ".p2align\t" . (log($$line)/log(2));

  536. 		} elsif ($dir eq ".section") {

  537. 		    $current_segment=$$line;

  538. 		    if (!$elf && $current_segment eq ".init") {

  539. 			if	($flavour eq "macosx")	{ $self->{value} = ".mod_init_func"; }

  540. 			elsif	($flavour eq "mingw64")	{ $self->{value} = ".section\t.ctors"; }

  541. 		    }

  542. 		} elsif ($dir =~ /\.(text|data)/) {

  543. 		    $current_segment=".$1";

  544. 		} elsif ($dir =~ /\.global|\.globl|\.extern/) {

  545. 		    if ($flavour eq "macosx") {

  546. 		        $self->{value} .= "\n.private_extern $$line";

  547. 		    } else {

  548. 		        $self->{value} .= "\n.hidden $$line";

  549. 		    }

  550. 		} elsif ($dir =~ /\.hidden/) {

  551. 		    if    ($flavour eq "macosx")  { $self->{value} = ".private_extern\t$prefix$$line"; }

  552. 		    elsif ($flavour eq "mingw64") { $self->{value} = ""; }

  553. 		} elsif ($dir =~ /\.comm/) {

  554. 		    $self->{value} = "$dir\t$prefix$$line";

  555. 		    $self->{value} =~ s|,([0-9]+),([0-9]+)$|",$1,".log($2)/log(2)|e if ($flavour eq "macosx");

  556. 		}

  557. 		$$line = "";

  558. 		return $self;

  559. 	    }

  560. 

  561. 	    # non-gas case or nasm/masm

  562. 	    SWITCH: for ($dir) {

  563. 		/\.text/    && do { my $v=undef;

  564. 				    if ($nasm) {

  565. 					$v="section	.text code align=64\n";

  566. 				    } else {

  567. 					$v="$current_segment\tENDS\n" if ($current_segment);

  568. 					$current_segment = ".text\$";

  569. 					$v.="$current_segment\tSEGMENT ";

  570. 					$v.=$masm>=$masmref ? "ALIGN(256)" : "PAGE";

  571. 					$v.=" 'CODE'";

  572. 				    }

  573. 				    $self->{value} = $v;

  574. 				    last;

  575. 				  };

  576. 		/\.data/    && do { my $v=undef;

  577. 				    if ($nasm) {

  578. 					$v="section	.data data align=8\n";

  579. 				    } else {

  580. 					$v="$current_segment\tENDS\n" if ($current_segment);

  581. 					$current_segment = "_DATA";

  582. 					$v.="$current_segment\tSEGMENT";

  583. 				    }

  584. 				    $self->{value} = $v;

  585. 				    last;

  586. 				  };

  587. 		/\.section/ && do { my $v=undef;

  588. 				    $$line =~ s/([^,]*).*/$1/;

  589. 				    $$line = ".CRT\$XCU" if ($$line eq ".init");

  590. 				    if ($nasm) {

  591. 					$v="section	$$line";

  592. 					if ($$line=~/\.([px])data/) {

  593. 					    $v.=" rdata align=";

  594. 					    $v.=$1 eq "p"? 4 : 8;

  595. 					} elsif ($$line=~/\.CRT\$/i) {

  596. 					    $v.=" rdata align=8";

  597. 					}

  598. 				    } else {

  599. 					$v="$current_segment\tENDS\n" if ($current_segment);

  600. 					$v.="$$line\tSEGMENT";

  601. 					if ($$line=~/\.([px])data/) {

  602. 					    $v.=" READONLY";

  603. 					    $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref);

  604. 					} elsif ($$line=~/\.CRT\$/i) {

  605. 					    $v.=" READONLY ";

  606. 					    $v.=$masm>=$masmref ? "ALIGN(8)" : "DWORD";

  607. 					}

  608. 				    }

  609. 				    $current_segment = $$line;

  610. 				    $self->{value} = $v;

  611. 				    last;

  612. 				  };

  613. 		/\.extern/  && do { $self->{value}  = "EXTERN\t".$$line;

  614. 				    $self->{value} .= ":NEAR" if ($masm);

  615. 				    last;

  616. 				  };

  617. 		/\.globl|.global/

  618. 			    && do { $self->{value}  = $masm?"PUBLIC":"global";

  619. 				    $self->{value} .= "\t".$$line;

  620. 				    last;

  621. 				  };

  622. 		/\.size/    && do { if (defined($current_function)) {

  623. 					undef $self->{value};

  624. 					if ($current_function->{abi} eq "svr4") {

  625. 					    $self->{value}="${decor}SEH_end_$current_function->{name}:";

  626. 					    $self->{value}.=":\n" if($masm);

  627. 					}

  628. 					$self->{value}.="$current_function->{name}\tENDP" if($masm && $current_function->{name});

  629. 					undef $current_function;

  630. 				    }

  631. 				    last;

  632. 				  };

  633. 		/\.align/   && do { my $max = ($masm && $masm>=$masmref) ? 256 : 4096;

  634. 				    $self->{value} = "ALIGN\t".($$line>$max?$max:$$line);

  635. 				    last;

  636. 				  };

  637. 		/\.(value|long|rva|quad)/

  638. 			    && do { my $sz  = substr($1,0,1);

  639. 				    my @arr = split(/,\s*/,$$line);

  640. 				    my $last = pop(@arr);

  641. 				    my $conv = sub  {	my $var=shift;

  642. 							$var=~s/^(0b[0-1]+)/oct($1)/eig;

  643. 							$var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm);

  644. 							if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva"))

  645. 							{ $var=~s/([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; }

  646. 							$var;

  647. 						    };  

  648. 

  649. 				    $sz =~ tr/bvlrq/BWDDQ/;

  650. 				    $self->{value} = "\tD$sz\t";

  651. 				    for (@arr) { $self->{value} .= &$conv($_).","; }

  652. 				    $self->{value} .= &$conv($last);

  653. 				    last;

  654. 				  };

  655. 		/\.byte/    && do { my @str=split(/,\s*/,$$line);

  656. 				    map(s/(0b[0-1]+)/oct($1)/eig,@str);

  657. 				    map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm);	

  658. 				    while ($#str>15) {

  659. 					$self->{value}.="DB\t"

  660. 						.join(",",@str[0..15])."\n";

  661. 					foreach (0..15) { shift @str; }

  662. 				    }

  663. 				    $self->{value}.="DB\t"

  664. 						.join(",",@str) if (@str);

  665. 				    last;

  666. 				  };

  667. 		/\.comm/    && do { my @str=split(/,\s*/,$$line);

  668. 				    my $v=undef;

  669. 				    if ($nasm) {

  670. 					$v.="common	$prefix@str[0] @str[1]";

  671. 				    } else {

  672. 					$v="$current_segment\tENDS\n" if ($current_segment);

  673. 					$current_segment = "_DATA";

  674. 					$v.="$current_segment\tSEGMENT\n";

  675. 					$v.="COMM	@str[0]:DWORD:".@str[1]/4;

  676. 				    }

  677. 				    $self->{value} = $v;

  678. 				    last;

  679. 				  };

  680. 	    }

  681. 	    $$line = "";

  682. 	}

  683. 

  684. 	$ret;

  685.     }

  686.     sub out {

  687. 	my $self = shift;

  688. 	$self->{value};

  689.     }

  690. }

  691. 

  692. sub rex {

  693.  my $opcode=shift;

  694.  my ($dst,$src,$rex)=@_;

  695. 

  696.    $rex|=0x04 if($dst>=8);

  697.    $rex|=0x01 if($src>=8);

  698.    push @$opcode,($rex|0x40) if ($rex);

  699. }

  700. 

  701. # Upon initial x86_64 introduction SSE>2 extensions were not introduced

  702. # yet. In order not to be bothered by tracing exact assembler versions,

  703. # but at the same time to provide a bare security minimum of AES-NI, we

  704. # hard-code some instructions. Extensions past AES-NI on the other hand

  705. # are traced by examining assembler version in individual perlasm

  706. # modules...

  707. 

  708. my %regrm = (	"%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3,

  709. 		"%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7	);

  710. 

  711. my $movq = sub {	# elderly gas can't handle inter-register movq

  712.   my $arg = shift;

  713.   my @opcode=(0x66);

  714.     if ($arg =~ /%xmm([0-9]+),\s*%r(\w+)/) {

  715. 	my ($src,$dst)=($1,$2);

  716. 	if ($dst !~ /[0-9]+/)	{ $dst = $regrm{"%e$dst"}; }

  717. 	rex(\@opcode,$src,$dst,0x8);

  718. 	push @opcode,0x0f,0x7e;

  719. 	push @opcode,0xc0|(($src&7)<<3)|($dst&7);	# ModR/M

  720. 	@opcode;

  721.     } elsif ($arg =~ /%r(\w+),\s*%xmm([0-9]+)/) {

  722. 	my ($src,$dst)=($2,$1);

  723. 	if ($dst !~ /[0-9]+/)	{ $dst = $regrm{"%e$dst"}; }

  724. 	rex(\@opcode,$src,$dst,0x8);

  725. 	push @opcode,0x0f,0x6e;

  726. 	push @opcode,0xc0|(($src&7)<<3)|($dst&7);	# ModR/M

  727. 	@opcode;

  728.     } else {

  729. 	();

  730.     }

  731. };

  732. 

  733. my $pextrd = sub {

  734.     if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*(%\w+)/) {

  735.       my @opcode=(0x66);

  736. 	my $imm=$1;

  737. 	my $src=$2;

  738. 	my $dst=$3;

  739. 	if ($dst =~ /%r([0-9]+)d/)	{ $dst = $1; }

  740. 	elsif ($dst =~ /%e/)		{ $dst = $regrm{$dst}; }

  741. 	rex(\@opcode,$src,$dst);

  742. 	push @opcode,0x0f,0x3a,0x16;

  743. 	push @opcode,0xc0|(($src&7)<<3)|($dst&7);	# ModR/M

  744. 	push @opcode,$imm;

  745. 	@opcode;

  746.     } else {

  747. 	();

  748.     }

  749. };

  750. 

  751. my $pinsrd = sub {

  752.     if (shift =~ /\$([0-9]+),\s*(%\w+),\s*%xmm([0-9]+)/) {

  753.       my @opcode=(0x66);

  754. 	my $imm=$1;

  755. 	my $src=$2;

  756. 	my $dst=$3;

  757. 	if ($src =~ /%r([0-9]+)/)	{ $src = $1; }

  758. 	elsif ($src =~ /%e/)		{ $src = $regrm{$src}; }

  759. 	rex(\@opcode,$dst,$src);

  760. 	push @opcode,0x0f,0x3a,0x22;

  761. 	push @opcode,0xc0|(($dst&7)<<3)|($src&7);	# ModR/M

  762. 	push @opcode,$imm;

  763. 	@opcode;

  764.     } else {

  765. 	();

  766.     }

  767. };

  768. 

  769. my $pshufb = sub {

  770.     if (shift =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) {

  771.       my @opcode=(0x66);

  772. 	rex(\@opcode,$2,$1);

  773. 	push @opcode,0x0f,0x38,0x00;

  774. 	push @opcode,0xc0|($1&7)|(($2&7)<<3);		# ModR/M

  775. 	@opcode;

  776.     } else {

  777. 	();

  778.     }

  779. };

  780. 

  781. my $palignr = sub {

  782.     if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {

  783.       my @opcode=(0x66);

  784. 	rex(\@opcode,$3,$2);

  785. 	push @opcode,0x0f,0x3a,0x0f;

  786. 	push @opcode,0xc0|($2&7)|(($3&7)<<3);		# ModR/M

  787. 	push @opcode,$1;

  788. 	@opcode;

  789.     } else {

  790. 	();

  791.     }

  792. };

  793. 

  794. my $pclmulqdq = sub {

  795.     if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {

  796.       my @opcode=(0x66);

  797. 	rex(\@opcode,$3,$2);

  798. 	push @opcode,0x0f,0x3a,0x44;

  799. 	push @opcode,0xc0|($2&7)|(($3&7)<<3);		# ModR/M

  800. 	my $c=$1;

  801. 	push @opcode,$c=~/^0/?oct($c):$c;

  802. 	@opcode;

  803.     } else {

  804. 	();

  805.     }

  806. };

  807. 

  808. my $rdrand = sub {

  809.     if (shift =~ /%[er](\w+)/) {

  810.       my @opcode=();

  811.       my $dst=$1;

  812. 	if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }

  813. 	rex(\@opcode,0,$1,8);

  814. 	push @opcode,0x0f,0xc7,0xf0|($dst&7);

  815. 	@opcode;

  816.     } else {

  817. 	();

  818.     }

  819. };

  820. 

  821. my $rdseed = sub {

  822.     if (shift =~ /%[er](\w+)/) {

  823.       my @opcode=();

  824.       my $dst=$1;

  825. 	if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }

  826. 	rex(\@opcode,0,$1,8);

  827. 	push @opcode,0x0f,0xc7,0xf8|($dst&7);

  828. 	@opcode;

  829.     } else {

  830. 	();

  831.     }

  832. };

  833. 

  834. sub rxb {

  835.  my $opcode=shift;

  836.  my ($dst,$src1,$src2,$rxb)=@_;

  837. 

  838.    $rxb|=0x7<<5;

  839.    $rxb&=~(0x04<<5) if($dst>=8);

  840.    $rxb&=~(0x01<<5) if($src1>=8);

  841.    $rxb&=~(0x02<<5) if($src2>=8);

  842.    push @$opcode,$rxb;

  843. }

  844. 

  845. my $vprotd = sub {

  846.     if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {

  847.       my @opcode=(0x8f);

  848. 	rxb(\@opcode,$3,$2,-1,0x08);

  849. 	push @opcode,0x78,0xc2;

  850. 	push @opcode,0xc0|($2&7)|(($3&7)<<3);		# ModR/M

  851. 	my $c=$1;

  852. 	push @opcode,$c=~/^0/?oct($c):$c;

  853. 	@opcode;

  854.     } else {

  855. 	();

  856.     }

  857. };

  858. 

  859. my $vprotq = sub {

  860.     if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {

  861.       my @opcode=(0x8f);

  862. 	rxb(\@opcode,$3,$2,-1,0x08);

  863. 	push @opcode,0x78,0xc3;

  864. 	push @opcode,0xc0|($2&7)|(($3&7)<<3);		# ModR/M

  865. 	my $c=$1;

  866. 	push @opcode,$c=~/^0/?oct($c):$c;

  867. 	@opcode;

  868.     } else {

  869. 	();

  870.     }

  871. };

  872. 

  873. my $endbranch = sub {

  874.     (0xf3,0x0f,0x1e,0xfa);

  875. };

  876. 

  877. if ($nasm) {

  878.     print <<___;

  879. default	rel

  880. %define XMMWORD

  881. %define YMMWORD

  882. %define ZMMWORD

  883. ___

  884. } elsif ($masm) {

  885.     print <<___;

  886. OPTION	DOTNAME

  887. ___

  888. }

  889. print STDOUT "#if defined(__x86_64__)\n" if ($gas);

  890. 

  891. while(defined(my $line=<>)) {

  892. 

  893.     $line =~ s|\R$||;           # Better chomp

  894. 

  895.     $line =~ s|[#!].*$||;	# get rid of asm-style comments...

  896.     $line =~ s|/\*.*\*/||;	# ... and C-style comments...

  897.     $line =~ s|^\s+||;		# ... and skip white spaces in beginning

  898.     $line =~ s|\s+$||;		# ... and at the end

  899. 

  900.     if (my $label=label->re(\$line))	{ print $label->out(); }

  901. 

  902.     if (my $directive=directive->re(\$line)) {

  903. 	printf "%s",$directive->out();

  904.     } elsif (my $opcode=opcode->re(\$line)) {

  905. 	my $asm = eval("\$".$opcode->mnemonic());

  906. 	

  907. 	if ((ref($asm) eq 'CODE') && scalar(my @bytes=&$asm($line))) {

  908. 	    print $gas?".byte\t":"DB\t",join(',',@bytes),"\n";

  909. 	    next;

  910. 	}

  911. 

  912. 	my @args;

  913. 	ARGUMENT: while (1) {

  914. 	    my $arg;

  915. 

  916. 	    ($arg=register->re(\$line, $opcode))||

  917. 	    ($arg=const->re(\$line))		||

  918. 	    ($arg=ea->re(\$line, $opcode))	||

  919. 	    ($arg=expr->re(\$line, $opcode))	||

  920. 	    last ARGUMENT;

  921. 

  922. 	    push @args,$arg;

  923. 

  924. 	    last ARGUMENT if ($line !~ /^,/);

  925. 

  926. 	    $line =~ s/^,\s*//;

  927. 	} # ARGUMENT:

  928. 

  929. 	if ($#args>=0) {

  930. 	    my $insn;

  931. 	    my $sz=$opcode->size();

  932. 

  933. 	    if ($gas) {

  934. 		$insn = $opcode->out($#args>=1?$args[$#args]->size():$sz);

  935. 		@args = map($_->out($sz),@args);

  936. 		printf "\t%s\t%s",$insn,join(",",@args);

  937. 	    } else {

  938. 		$insn = $opcode->out();

  939. 		foreach (@args) {

  940. 		    my $arg = $_->out();

  941. 		    # $insn.=$sz compensates for movq, pinsrw, ...

  942. 		    if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; }

  943. 		    if ($arg =~ /^ymm[0-9]+$/) { $insn.=$sz; $sz="y" if(!$sz); last; }

  944. 		    if ($arg =~ /^zmm[0-9]+$/) { $insn.=$sz; $sz="z" if(!$sz); last; }

  945. 		    if ($arg =~ /^mm[0-9]+$/)  { $insn.=$sz; $sz="q" if(!$sz); last; }

  946. 		}

  947. 		@args = reverse(@args);

  948. 		undef $sz if ($nasm && $opcode->mnemonic() eq "lea");

  949. 

  950. 		if ($insn eq "movq" && $#args == 1 && $args[0]->out($sz) eq "xmm0" && $args[1]->out($sz) eq "rax") {

  951. 		    # I have no clue why MASM can't parse this instruction.

  952. 		    printf "DB 66h, 48h, 0fh, 6eh, 0c0h";

  953. 		} else {

  954. 		    printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args));

  955. 		}

  956. 	    }

  957. 	} else {

  958. 	    printf "\t%s",$opcode->out();

  959. 	}

  960.     }

  961. 

  962.     print $line,"\n";

  963. }

  964. 

  965. print "\n$current_segment\tENDS\n"	if ($current_segment && $masm);

  966. print "END\n"				if ($masm);

  967. print "#endif\n"			if ($gas);

  968. 

  969. 

  970. close STDOUT;

  971. 

  972. #################################################

  973. # Cross-reference x86_64 ABI "card"

  974. #

  975. # 		Unix		Win64

  976. # %rax		*		*

  977. # %rbx		-		-

  978. # %rcx		#4		#1

  979. # %rdx		#3		#2

  980. # %rsi		#2		-

  981. # %rdi		#1		-

  982. # %rbp		-		-

  983. # %rsp		-		-

  984. # %r8		#5		#3

  985. # %r9		#6		#4

  986. # %r10		*		*

  987. # %r11		*		*

  988. # %r12		-		-

  989. # %r13		-		-

  990. # %r14		-		-

  991. # %r15		-		-

  992. # 

  993. # (*)	volatile register

  994. # (-)	preserved by callee

  995. # (#)	Nth argument, volatile

  996. #

  997. # In Unix terms top of stack is argument transfer area for arguments

  998. # which could not be accommodated in registers. Or in other words 7th

  999. # [integer] argument resides at 8(%rsp) upon function entry point.

  1000. # 128 bytes above %rsp constitute a "red zone" which is not touched

  1001. # by signal handlers and can be used as temporal storage without

  1002. # allocating a frame.

  1003. #

  1004. # In Win64 terms N*8 bytes on top of stack is argument transfer area,

  1005. # which belongs to/can be overwritten by callee. N is the number of

  1006. # arguments passed to callee, *but* not less than 4! This means that

  1007. # upon function entry point 5th argument resides at 40(%rsp), as well

  1008. # as that 32 bytes from 8(%rsp) can always be used as temporal

  1009. # storage [without allocating a frame]. One can actually argue that

  1010. # one can assume a "red zone" above stack pointer under Win64 as well.

  1011. # Point is that at apparently no occasion Windows kernel would alter

  1012. # the area above user stack pointer in true asynchronous manner...

  1013. #

  1014. # All the above means that if assembler programmer adheres to Unix

  1015. # register and stack layout, but disregards the "red zone" existense,

  1016. # it's possible to use following prologue and epilogue to "gear" from

  1017. # Unix to Win64 ABI in leaf functions with not more than 6 arguments.

  1018. #

  1019. # omnipotent_function:

  1020. # ifdef WIN64

  1021. #	movq	%rdi,8(%rsp)

  1022. #	movq	%rsi,16(%rsp)

  1023. #	movq	%rcx,%rdi	; if 1st argument is actually present

  1024. #	movq	%rdx,%rsi	; if 2nd argument is actually ...

  1025. #	movq	%r8,%rdx	; if 3rd argument is ...

  1026. #	movq	%r9,%rcx	; if 4th argument ...

  1027. #	movq	40(%rsp),%r8	; if 5th ...

  1028. #	movq	48(%rsp),%r9	; if 6th ...

  1029. # endif

  1030. #	...

  1031. # ifdef WIN64

  1032. #	movq	8(%rsp),%rdi

  1033. #	movq	16(%rsp),%rsi

  1034. # endif

  1035. #	ret

  1036. #

  1037. #################################################

  1038. # Win64 SEH, Structured Exception Handling.

  1039. #

  1040. # Unlike on Unix systems(*) lack of Win64 stack unwinding information

  1041. # has undesired side-effect at run-time: if an exception is raised in

  1042. # assembler subroutine such as those in question (basically we're

  1043. # referring to segmentation violations caused by malformed input

  1044. # parameters), the application is briskly terminated without invoking

  1045. # any exception handlers, most notably without generating memory dump

  1046. # or any user notification whatsoever. This poses a problem. It's

  1047. # possible to address it by registering custom language-specific

  1048. # handler that would restore processor context to the state at

  1049. # subroutine entry point and return "exception is not handled, keep

  1050. # unwinding" code. Writing such handler can be a challenge... But it's

  1051. # doable, though requires certain coding convention. Consider following

  1052. # snippet:

  1053. #

  1054. # .type	function,@function

  1055. # function:

  1056. #	movq	%rsp,%rax	# copy rsp to volatile register

  1057. #	pushq	%r15		# save non-volatile registers

  1058. #	pushq	%rbx

  1059. #	pushq	%rbp

  1060. #	movq	%rsp,%r11

  1061. #	subq	%rdi,%r11	# prepare [variable] stack frame

  1062. #	andq	$-64,%r11

  1063. #	movq	%rax,0(%r11)	# check for exceptions

  1064. #	movq	%r11,%rsp	# allocate [variable] stack frame

  1065. #	movq	%rax,0(%rsp)	# save original rsp value

  1066. # magic_point:

  1067. #	...

  1068. #	movq	0(%rsp),%rcx	# pull original rsp value

  1069. #	movq	-24(%rcx),%rbp	# restore non-volatile registers

  1070. #	movq	-16(%rcx),%rbx

  1071. #	movq	-8(%rcx),%r15

  1072. #	movq	%rcx,%rsp	# restore original rsp

  1073. #	ret

  1074. # .size function,.-function

  1075. #

  1076. # The key is that up to magic_point copy of original rsp value remains

  1077. # in chosen volatile register and no non-volatile register, except for

  1078. # rsp, is modified. While past magic_point rsp remains constant till

  1079. # the very end of the function. In this case custom language-specific

  1080. # exception handler would look like this:

  1081. #

  1082. # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,

  1083. #		CONTEXT *context,DISPATCHER_CONTEXT *disp)

  1084. # {	ULONG64 *rsp = (ULONG64 *)context->Rax;

  1085. #	if (context->Rip >= magic_point)

  1086. #	{   rsp = ((ULONG64 **)context->Rsp)[0];

  1087. #	    context->Rbp = rsp[-3];

  1088. #	    context->Rbx = rsp[-2];

  1089. #	    context->R15 = rsp[-1];

  1090. #	}

  1091. #	context->Rsp = (ULONG64)rsp;

  1092. #	context->Rdi = rsp[1];

  1093. #	context->Rsi = rsp[2];

  1094. #

  1095. #	memcpy (disp->ContextRecord,context,sizeof(CONTEXT));

  1096. #	RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase,

  1097. #		dips->ControlPc,disp->FunctionEntry,disp->ContextRecord,

  1098. #		&disp->HandlerData,&disp->EstablisherFrame,NULL);

  1099. #	return ExceptionContinueSearch;

  1100. # }

  1101. #

  1102. # It's appropriate to implement this handler in assembler, directly in

  1103. # function's module. In order to do that one has to know members'

  1104. # offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant

  1105. # values. Here they are:

  1106. #

  1107. #	CONTEXT.Rax				120

  1108. #	CONTEXT.Rcx				128

  1109. #	CONTEXT.Rdx				136

  1110. #	CONTEXT.Rbx				144

  1111. #	CONTEXT.Rsp				152

  1112. #	CONTEXT.Rbp				160

  1113. #	CONTEXT.Rsi				168

  1114. #	CONTEXT.Rdi				176

  1115. #	CONTEXT.R8				184

  1116. #	CONTEXT.R9				192

  1117. #	CONTEXT.R10				200

  1118. #	CONTEXT.R11				208

  1119. #	CONTEXT.R12				216

  1120. #	CONTEXT.R13				224

  1121. #	CONTEXT.R14				232

  1122. #	CONTEXT.R15				240

  1123. #	CONTEXT.Rip				248

  1124. #	CONTEXT.Xmm6				512

  1125. #	sizeof(CONTEXT)				1232

  1126. #	DISPATCHER_CONTEXT.ControlPc		0

  1127. #	DISPATCHER_CONTEXT.ImageBase		8

  1128. #	DISPATCHER_CONTEXT.FunctionEntry	16

  1129. #	DISPATCHER_CONTEXT.EstablisherFrame	24

  1130. #	DISPATCHER_CONTEXT.TargetIp		32

  1131. #	DISPATCHER_CONTEXT.ContextRecord	40

  1132. #	DISPATCHER_CONTEXT.LanguageHandler	48

  1133. #	DISPATCHER_CONTEXT.HandlerData		56

  1134. #	UNW_FLAG_NHANDLER			0

  1135. #	ExceptionContinueSearch			1

  1136. #

  1137. # In order to tie the handler to the function one has to compose

  1138. # couple of structures: one for .xdata segment and one for .pdata.

  1139. #

  1140. # UNWIND_INFO structure for .xdata segment would be

  1141. #

  1142. # function_unwind_info:

  1143. #	.byte	9,0,0,0

  1144. #	.rva	handler

  1145. #

  1146. # This structure designates exception handler for a function with

  1147. # zero-length prologue, no stack frame or frame register.

  1148. #

  1149. # To facilitate composing of .pdata structures, auto-generated "gear"

  1150. # prologue copies rsp value to rax and denotes next instruction with

  1151. # .LSEH_begin_{function_name} label. This essentially defines the SEH

  1152. # styling rule mentioned in the beginning. Position of this label is

  1153. # chosen in such manner that possible exceptions raised in the "gear"

  1154. # prologue would be accounted to caller and unwound from latter's frame.

  1155. # End of function is marked with respective .LSEH_end_{function_name}

  1156. # label. To summarize, .pdata segment would contain

  1157. #

  1158. #	.rva	.LSEH_begin_function

  1159. #	.rva	.LSEH_end_function

  1160. #	.rva	function_unwind_info

  1161. #

  1162. # Reference to function_unwind_info from .xdata segment is the anchor.

  1163. # In case you wonder why references are 32-bit .rvas and not 64-bit

  1164. # .quads. References put into these two segments are required to be

  1165. # *relative* to the base address of the current binary module, a.k.a.

  1166. # image base. No Win64 module, be it .exe or .dll, can be larger than

  1167. # 2GB and thus such relative references can be and are accommodated in

  1168. # 32 bits.

  1169. #

  1170. # Having reviewed the example function code, one can argue that "movq

  1171. # %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix

  1172. # rax would contain an undefined value. If this "offends" you, use

  1173. # another register and refrain from modifying rax till magic_point is

  1174. # reached, i.e. as if it was a non-volatile register. If more registers

  1175. # are required prior [variable] frame setup is completed, note that

  1176. # nobody says that you can have only one "magic point." You can

  1177. # "liberate" non-volatile registers by denoting last stack off-load

  1178. # instruction and reflecting it in finer grade unwind logic in handler.

  1179. # After all, isn't it why it's called *language-specific* handler...

  1180. #

  1181. # Attentive reader can notice that exceptions would be mishandled in

  1182. # auto-generated "gear" epilogue. Well, exception effectively can't

  1183. # occur there, because if memory area used by it was subject to

  1184. # segmentation violation, then it would be raised upon call to the

  1185. # function (and as already mentioned be accounted to caller, which is

  1186. # not a problem). If you're still not comfortable, then define tail

  1187. # "magic point" just prior ret instruction and have handler treat it...

  1188. #

  1189. # (*)	Note that we're talking about run-time, not debug-time. Lack of

  1190. #	unwind information makes debugging hard on both Windows and

  1191. #	Unix. "Unlike" referes to the fact that on Unix signal handler

  1192. #	will always be invoked, core dumped and appropriate exit code

  1193. #	returned to parent (for user notification).