kris
/
boringssl
1
0
Fork 0
Code Issues 0 Pull Requests 0 Releases 0 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
841 Commits
12 Branches
38 MiB
 
 
 
 
 
 
Tree: 78e6978ab9
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '78e6978ab9'
${ noResults }
boringssl/crypto/perlasm/x86_64-xlate.pl

1166 lines
35 KiB
Raw Blame History 

  1. #!/usr/bin/env perl

  2. 

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

  4. #

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

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

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

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

  9. #

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

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

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

  13. #

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

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

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

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

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

  19. #

  20. # Currently recognized limitations:

  21. #

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

  23. #

  24. # Dual-ABI styling rules.

  25. #

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  40. #    (sorry about latter).

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

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

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

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

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

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

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

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

  49. #    target register, e.g.:

  50. #

  51. #		.picmeup	%rax

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

  53. #

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

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

  56. #    see SEH paragraph at the end.

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

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

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

  60. 

  61. my $flavour = shift;

  62. my $output  = shift;

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

  64. 

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

  66. 	if (defined($output));

  67. 

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

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

  70. my $win64=0;

  71. my $prefix="";

  72. my $decor=".L";

  73. 

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

  75. my $masm=0;

  76. my $PTR=" PTR";

  77. 

  78. my $nasmref=2.03;

  79. my $nasm=0;

  80. 

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

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

  83. 				  chomp($prefix);

  84. 				}

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

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

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

  88. elsif (!$gas)

  89. {   if ($ENV{ASM} =~ m/nasm/ && `nasm -v` =~ m/version ([0-9]+)\.([0-9]+)/i)

  90.     {	$nasm = $1 + $2*0.01; $PTR="";  }

  91.     elsif (`ml64 2>&1` =~ m/Version ([0-9]+)\.([0-9]+)(\.([0-9]+))?/)

  92.     {	$masm = $1 + $2*2**-16 + $4*2**-32;   }

  93.     die "no assembler found on %PATH" if (!($nasm || $masm));

  94.     $win64=1;

  95.     $elf=0;

  96.     $decor="\$L\$";

  97. }

  98. 

  99. my $current_segment;

  100. my $current_function;

  101. my %globals;

  102. 

  103. { package opcode;	# pick up opcodes

  104.     sub re {

  105. 	my	$self = shift;	# single instance in enough...

  106. 	local	*line = shift;

  107. 	undef	$ret;

  108. 

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

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

  111. 	    $ret = $self;

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

  113. 

  114. 	    undef $self->{sz};

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  129. 	    }

  130. 	}

  131. 	$ret;

  132.     }

  133.     sub size {

  134. 	my $self = shift;

  135. 	my $sz   = shift;

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

  137. 	$self->{sz};

  138.     }

  139.     sub out {

  140. 	my $self = shift;

  141. 	if ($gas) {

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

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

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

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

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

  147. 		my $epilogue = "";

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

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

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

  151. 		}

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

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

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

  155. 	    } else {

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

  157. 	    }

  158. 	} else {

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

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

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

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

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

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

  165. 	    	}

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

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

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

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

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

  171. 	    } 

  172. 	    $self->{op};

  173. 	}

  174.     }

  175.     sub mnemonic {

  176. 	my $self=shift;

  177. 	my $op=shift;

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

  179. 	$self->{op};

  180.     }

  181. }

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

  183.     sub re {

  184. 	my	$self = shift;	# single instance in enough...

  185. 	local	*line = shift;

  186. 	undef	$ret;

  187. 

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

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

  190. 	    $ret = $self;

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

  192. 	}

  193. 	$ret;

  194.     }

  195.     sub out {

  196.     	my $self = shift;

  197. 

  198. 	if ($gas) {

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

  200. 	    # in $self->{value}

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

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

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

  204. 	} else {

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

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

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

  208. 	}

  209.     }

  210. }

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

  212.     sub re {

  213. 	my	$self = shift;	# single instance in enough...

  214. 	local	*line = shift;

  215. 	undef	$ret;

  216. 

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

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

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

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

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

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

  223. 	    $ret = $self;

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

  225. 

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

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

  228. 		opcode->mnemonic("lea");

  229. 	    }

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

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

  232. 	}

  233. 	$ret;

  234.     }

  235.     sub size {}

  236.     sub out {

  237.     	my $self = shift;

  238. 	my $sz = shift;

  239. 

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

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

  242. 

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

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

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

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

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

  248. 

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

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

  251. 	use integer;

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

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

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

  255. 

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

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

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

  259. 	}

  260. 

  261. 	if ($gas) {

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

  263. 

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

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

  266. 					$self->{label},

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

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

  269. 	    } else {

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

  271. 	    }

  272. 	} else {

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

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

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

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

  277. 

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

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

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

  281. 

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

  283. 	    (opcode->mnemonic() =~ /^v?mov([qd])$/)		&& ($sz=$1)  ||

  284. 	    (opcode->mnemonic() =~ /^v?pinsr([qdwb])$/)		&& ($sz=$1)  ||

  285. 	    (opcode->mnemonic() =~ /^vpbroadcast([qdwb])$/)	&& ($sz=$1)  ||

  286. 	    (opcode->mnemonic() =~ /^vinsert[fi]128$/)		&& ($sz="x");

  287. 

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

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

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

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

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

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

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

  295. 	    } else {

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

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

  298. 					$self->{base};

  299. 	    }

  300. 	}

  301.     }

  302. }

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

  304.     sub re {

  305. 	my	$class = shift;	# muliple instances...

  306. 	my	$self = {};

  307. 	local	*line = shift;

  308. 	undef	$ret;

  309. 

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

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

  312. 	    bless $self,$class;

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

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

  315. 	    $ret = $self;

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

  317. 	}

  318. 	$ret;

  319.     }

  320.     sub size {

  321. 	my	$self = shift;

  322. 	undef	$ret;

  323. 

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

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

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

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

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

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

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

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

  332. 

  333. 	$ret;

  334.     }

  335.     sub out {

  336.     	my $self = shift;

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

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

  339.     }

  340. }

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

  342.     sub re {

  343. 	my	$self = shift;	# single instance is enough...

  344. 	local	*line = shift;

  345. 	undef	$ret;

  346. 

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

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

  349. 	    $ret = $self;

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

  351. 

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

  353. 	}

  354. 	$ret;

  355.     }

  356.     sub out {

  357. 	my $self = shift;

  358. 

  359. 	if ($gas) {

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

  361. 	    if ($win64	&&

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

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

  364. 		$func .= "\n";

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

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

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

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

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

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

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

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

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

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

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

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

  377. 	    }

  378. 	    $func;

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

  380. 	    $self->{value} .= ":" if ($masm && $ret!~m/^\$/);

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

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

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

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

  385. 			"\n";

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

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

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

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

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

  391. 	    $func .= "\n";

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

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

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

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

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

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

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

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

  400. 	    $func .= "\n";

  401. 	} else {

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

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

  404. 	}

  405.     }

  406. }

  407. { package expr;		# pick up expressioins

  408.     sub re {

  409. 	my	$self = shift;	# single instance is enough...

  410. 	local	*line = shift;

  411. 	undef	$ret;

  412. 

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

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

  415. 	    $ret = $self;

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

  417. 

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

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

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

  421. 	}

  422. 	$ret;

  423.     }

  424.     sub out {

  425. 	my $self = shift;

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

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

  428. 	} else {

  429. 	    $self->{value};

  430. 	}

  431.     }

  432. }

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

  434.     sub re {

  435. 	my	$self = shift;	# single instance is enough...

  436. 	local	*line = shift;

  437. 	undef	$ret;

  438. 	my	$dir;

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

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

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

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

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

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

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

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

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

  448. 

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

  450. 	    $dir = $1;

  451. 	    $ret = $self;

  452. 	    undef $self->{value};

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

  454. 

  455. 	    SWITCH: for ($dir) {

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

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

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

  459. 				    }

  460. 				    last;

  461. 				  };

  462. 		/\.global|\.globl|\.extern/

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

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

  465. 				    last;

  466. 				  };

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

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

  469. 					undef $current_function;

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

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

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

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

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

  475. 					undef $current_function;

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

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

  478. 				    }

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

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

  481. 				    last;

  482. 				  };

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

  484. 					$dir  = ".byte";

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

  486. 				    }

  487. 				    last;

  488. 				  };

  489. 		/\.rva|\.long|\.quad/

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

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

  492. 				    last;

  493. 				  };

  494. 	    }

  495. 

  496. 	    if ($gas) {

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

  498. 

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

  500. 		    $self->{value} = ""; # swallow extern

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

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

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

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

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

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

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

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

  509. 		    if (defined($current_function)) {

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

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

  512. 			undef $current_function;

  513. 		    }

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

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

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

  517. 		    $current_segment=$line;

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

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

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

  521. 		    }

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

  523. 		    $current_segment=".$1";

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

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

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

  527. 		    } else {

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

  529. 		    }

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

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

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

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

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

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

  536. 		}

  537. 		$line = "";

  538. 		return $self;

  539. 	    }

  540. 

  541. 	    # non-gas case or nasm/masm

  542. 	    SWITCH: for ($dir) {

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

  544. 				    if ($nasm) {

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

  546. 				    } else {

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

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

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

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

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

  552. 				    }

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

  554. 				    last;

  555. 				  };

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

  557. 				    if ($nasm) {

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

  559. 				    } else {

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

  561. 					$current_segment = "_DATA";

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

  563. 				    }

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

  565. 				    last;

  566. 				  };

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

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

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

  570. 				    if ($nasm) {

  571. 					$v="section	$line";

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

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

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

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

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

  577. 					}

  578. 				    } else {

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

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

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

  582. 					    $v.=" READONLY";

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

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

  585. 					    $v.=" READONLY ";

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

  587. 					}

  588. 				    }

  589. 				    $current_segment = $line;

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

  591. 				    last;

  592. 				  };

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

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

  595. 				    last;

  596. 				  };

  597. 		/\.globl|.global/

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

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

  600. 				    last;

  601. 				  };

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

  603. 					undef $self->{value};

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

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

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

  607. 					}

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

  609. 					undef $current_function;

  610. 				    }

  611. 				    last;

  612. 				  };

  613. 		/\.align/   && do { $self->{value} = "ALIGN\t".$line; last; };

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

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

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

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

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

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

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

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

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

  623. 							$var;

  624. 						    };  

  625. 

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

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

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

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

  630. 				    last;

  631. 				  };

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

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

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

  635. 				    while ($#str>15) {

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

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

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

  639. 				    }

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

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

  642. 				    last;

  643. 				  };

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

  645. 				    my $v=undef;

  646. 				    if ($nasm) {

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

  648. 				    } else {

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

  650. 					$current_segment = "_DATA";

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

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

  653. 				    }

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

  655. 				    last;

  656. 				  };

  657. 	    }

  658. 	    $line = "";

  659. 	}

  660. 

  661. 	$ret;

  662.     }

  663.     sub out {

  664. 	my $self = shift;

  665. 	$self->{value};

  666.     }

  667. }

  668. 

  669. sub rex {

  670.  local *opcode=shift;

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

  672. 

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

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

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

  676. }

  677. 

  678. # older gas and ml64 don't handle SSE>2 instructions

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

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

  681. 

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

  683.   my $arg = shift;

  684.   my @opcode=(0x66);

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

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

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

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

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

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

  691. 	@opcode;

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

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

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

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

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

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

  698. 	@opcode;

  699.     } else {

  700. 	();

  701.     }

  702. };

  703. 

  704. my $pextrd = sub {

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

  706.       my @opcode=(0x66);

  707. 	$imm=$1;

  708. 	$src=$2;

  709. 	$dst=$3;

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

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

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

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

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

  715. 	push @opcode,$imm;

  716. 	@opcode;

  717.     } else {

  718. 	();

  719.     }

  720. };

  721. 

  722. my $pinsrd = sub {

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

  724.       my @opcode=(0x66);

  725. 	$imm=$1;

  726. 	$src=$2;

  727. 	$dst=$3;

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

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

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

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

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

  733. 	push @opcode,$imm;

  734. 	@opcode;

  735.     } else {

  736. 	();

  737.     }

  738. };

  739. 

  740. my $pshufb = sub {

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

  742.       my @opcode=(0x66);

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

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

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

  746. 	@opcode;

  747.     } else {

  748. 	();

  749.     }

  750. };

  751. 

  752. my $palignr = sub {

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

  754.       my @opcode=(0x66);

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

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

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

  758. 	push @opcode,$1;

  759. 	@opcode;

  760.     } else {

  761. 	();

  762.     }

  763. };

  764. 

  765. my $pclmulqdq = sub {

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

  767.       my @opcode=(0x66);

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

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

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

  771. 	my $c=$1;

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

  773. 	@opcode;

  774.     } else {

  775. 	();

  776.     }

  777. };

  778. 

  779. my $rdrand = sub {

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

  781.       my @opcode=();

  782.       my $dst=$1;

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

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

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

  786. 	@opcode;

  787.     } else {

  788. 	();

  789.     }

  790. };

  791. 

  792. my $rdseed = sub {

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

  794.       my @opcode=();

  795.       my $dst=$1;

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

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

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

  799. 	@opcode;

  800.     } else {

  801. 	();

  802.     }

  803. };

  804. 

  805. sub rxb {

  806.  local *opcode=shift;

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

  808. 

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

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

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

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

  813.    push @opcode,$rxb;

  814. }

  815. 

  816. my $vprotd = sub {

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

  818.       my @opcode=(0x8f);

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

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

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

  822. 	my $c=$1;

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

  824. 	@opcode;

  825.     } else {

  826. 	();

  827.     }

  828. };

  829. 

  830. my $vprotq = sub {

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

  832.       my @opcode=(0x8f);

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

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

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

  836. 	my $c=$1;

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

  838. 	@opcode;

  839.     } else {

  840. 	();

  841.     }

  842. };

  843. 

  844. if ($nasm) {

  845.     print <<___;

  846. default	rel

  847. %define XMMWORD

  848. %define YMMWORD

  849. %define ZMMWORD

  850. ___

  851. } elsif ($masm) {

  852.     print <<___;

  853. OPTION	DOTNAME

  854. ___

  855. }

  856. 

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

  858. 

  859. while($line=<>) {

  860. 

  861.     chomp($line);

  862. 

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

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

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

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

  867. 

  868.     undef $label;

  869.     undef $opcode;

  870.     undef @args;

  871. 

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

  873. 

  874.     if (directive->re(\$line)) {

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

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

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

  878. 	undef @bytes;

  879. 	

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

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

  882. 	    next;

  883. 	}

  884. 

  885. 	ARGUMENT: while (1) {

  886. 	my $arg;

  887. 

  888. 	if ($arg=register->re(\$line))	{ opcode->size($arg->size()); }

  889. 	elsif ($arg=const->re(\$line))	{ }

  890. 	elsif ($arg=ea->re(\$line))	{ }

  891. 	elsif ($arg=expr->re(\$line))	{ }

  892. 	else				{ last ARGUMENT; }

  893. 

  894. 	push @args,$arg;

  895. 

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

  897. 

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

  899. 	} # ARGUMENT:

  900. 

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

  902. 	    my $insn;

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

  904. 

  905. 	    if ($gas) {

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

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

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

  909. 	    } else {

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

  911. 		foreach (@args) {

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

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

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

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

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

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

  918. 		}

  919. 		@args = reverse(@args);

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

  921. 

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

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

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

  925. 		} else {

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

  927. 		}

  928. 	    }

  929. 	} else {

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

  931. 	}

  932.     }

  933. 

  934.     print $line,"\n";

  935. }

  936. 

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

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

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

  940. 

  941. 

  942. close STDOUT;

  943. 

  944. #################################################

  945. # Cross-reference x86_64 ABI "card"

  946. #

  947. # 		Unix		Win64

  948. # %rax		*		*

  949. # %rbx		-		-

  950. # %rcx		#4		#1

  951. # %rdx		#3		#2

  952. # %rsi		#2		-

  953. # %rdi		#1		-

  954. # %rbp		-		-

  955. # %rsp		-		-

  956. # %r8		#5		#3

  957. # %r9		#6		#4

  958. # %r10		*		*

  959. # %r11		*		*

  960. # %r12		-		-

  961. # %r13		-		-

  962. # %r14		-		-

  963. # %r15		-		-

  964. # 

  965. # (*)	volatile register

  966. # (-)	preserved by callee

  967. # (#)	Nth argument, volatile

  968. #

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

  970. # which could not be accomodated in registers. Or in other words 7th

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

  972. # 128 bytes above %rsp constitute a "red zone" which is not touched

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

  974. # allocating a frame.

  975. #

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

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

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

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

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

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

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

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

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

  985. #

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

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

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

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

  990. #

  991. # omnipotent_function:

  992. # ifdef WIN64

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

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

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

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

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

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

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

  1000. #	movq	48(%rsp),%r9	; if 6th ...

  1001. # endif

  1002. #	...

  1003. # ifdef WIN64

  1004. #	movq	8(%rsp),%rdi

  1005. #	movq	16(%rsp),%rsi

  1006. # endif

  1007. #	ret

  1008. #

  1009. #################################################

  1010. # Win64 SEH, Structured Exception Handling.

  1011. #

  1012. # Unlike on Unix systems(*) lack of Win64 stack unwinding information

  1013. # has undesired side-effect at run-time: if an exception is raised in

  1014. # assembler subroutine such as those in question (basically we're

  1015. # referring to segmentation violations caused by malformed input

  1016. # parameters), the application is briskly terminated without invoking

  1017. # any exception handlers, most notably without generating memory dump

  1018. # or any user notification whatsoever. This poses a problem. It's

  1019. # possible to address it by registering custom language-specific

  1020. # handler that would restore processor context to the state at

  1021. # subroutine entry point and return "exception is not handled, keep

  1022. # unwinding" code. Writing such handler can be a challenge... But it's

  1023. # doable, though requires certain coding convention. Consider following

  1024. # snippet:

  1025. #

  1026. # .type	function,@function

  1027. # function:

  1028. #	movq	%rsp,%rax	# copy rsp to volatile register

  1029. #	pushq	%r15		# save non-volatile registers

  1030. #	pushq	%rbx

  1031. #	pushq	%rbp

  1032. #	movq	%rsp,%r11

  1033. #	subq	%rdi,%r11	# prepare [variable] stack frame

  1034. #	andq	$-64,%r11

  1035. #	movq	%rax,0(%r11)	# check for exceptions

  1036. #	movq	%r11,%rsp	# allocate [variable] stack frame

  1037. #	movq	%rax,0(%rsp)	# save original rsp value

  1038. # magic_point:

  1039. #	...

  1040. #	movq	0(%rsp),%rcx	# pull original rsp value

  1041. #	movq	-24(%rcx),%rbp	# restore non-volatile registers

  1042. #	movq	-16(%rcx),%rbx

  1043. #	movq	-8(%rcx),%r15

  1044. #	movq	%rcx,%rsp	# restore original rsp

  1045. #	ret

  1046. # .size function,.-function

  1047. #

  1048. # The key is that up to magic_point copy of original rsp value remains

  1049. # in chosen volatile register and no non-volatile register, except for

  1050. # rsp, is modified. While past magic_point rsp remains constant till

  1051. # the very end of the function. In this case custom language-specific

  1052. # exception handler would look like this:

  1053. #

  1054. # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,

  1055. #		CONTEXT *context,DISPATCHER_CONTEXT *disp)

  1056. # {	ULONG64 *rsp = (ULONG64 *)context->Rax;

  1057. #	if (context->Rip >= magic_point)

  1058. #	{   rsp = ((ULONG64 **)context->Rsp)[0];

  1059. #	    context->Rbp = rsp[-3];

  1060. #	    context->Rbx = rsp[-2];

  1061. #	    context->R15 = rsp[-1];

  1062. #	}

  1063. #	context->Rsp = (ULONG64)rsp;

  1064. #	context->Rdi = rsp[1];

  1065. #	context->Rsi = rsp[2];

  1066. #

  1067. #	memcpy (disp->ContextRecord,context,sizeof(CONTEXT));

  1068. #	RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase,

  1069. #		dips->ControlPc,disp->FunctionEntry,disp->ContextRecord,

  1070. #		&disp->HandlerData,&disp->EstablisherFrame,NULL);

  1071. #	return ExceptionContinueSearch;

  1072. # }

  1073. #

  1074. # It's appropriate to implement this handler in assembler, directly in

  1075. # function's module. In order to do that one has to know members'

  1076. # offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant

  1077. # values. Here they are:

  1078. #

  1079. #	CONTEXT.Rax				120

  1080. #	CONTEXT.Rcx				128

  1081. #	CONTEXT.Rdx				136

  1082. #	CONTEXT.Rbx				144

  1083. #	CONTEXT.Rsp				152

  1084. #	CONTEXT.Rbp				160

  1085. #	CONTEXT.Rsi				168

  1086. #	CONTEXT.Rdi				176

  1087. #	CONTEXT.R8				184

  1088. #	CONTEXT.R9				192

  1089. #	CONTEXT.R10				200

  1090. #	CONTEXT.R11				208

  1091. #	CONTEXT.R12				216

  1092. #	CONTEXT.R13				224

  1093. #	CONTEXT.R14				232

  1094. #	CONTEXT.R15				240

  1095. #	CONTEXT.Rip				248

  1096. #	CONTEXT.Xmm6				512

  1097. #	sizeof(CONTEXT)				1232

  1098. #	DISPATCHER_CONTEXT.ControlPc		0

  1099. #	DISPATCHER_CONTEXT.ImageBase		8

  1100. #	DISPATCHER_CONTEXT.FunctionEntry	16

  1101. #	DISPATCHER_CONTEXT.EstablisherFrame	24

  1102. #	DISPATCHER_CONTEXT.TargetIp		32

  1103. #	DISPATCHER_CONTEXT.ContextRecord	40

  1104. #	DISPATCHER_CONTEXT.LanguageHandler	48

  1105. #	DISPATCHER_CONTEXT.HandlerData		56

  1106. #	UNW_FLAG_NHANDLER			0

  1107. #	ExceptionContinueSearch			1

  1108. #

  1109. # In order to tie the handler to the function one has to compose

  1110. # couple of structures: one for .xdata segment and one for .pdata.

  1111. #

  1112. # UNWIND_INFO structure for .xdata segment would be

  1113. #

  1114. # function_unwind_info:

  1115. #	.byte	9,0,0,0

  1116. #	.rva	handler

  1117. #

  1118. # This structure designates exception handler for a function with

  1119. # zero-length prologue, no stack frame or frame register.

  1120. #

  1121. # To facilitate composing of .pdata structures, auto-generated "gear"

  1122. # prologue copies rsp value to rax and denotes next instruction with

  1123. # .LSEH_begin_{function_name} label. This essentially defines the SEH

  1124. # styling rule mentioned in the beginning. Position of this label is

  1125. # chosen in such manner that possible exceptions raised in the "gear"

  1126. # prologue would be accounted to caller and unwound from latter's frame.

  1127. # End of function is marked with respective .LSEH_end_{function_name}

  1128. # label. To summarize, .pdata segment would contain

  1129. #

  1130. #	.rva	.LSEH_begin_function

  1131. #	.rva	.LSEH_end_function

  1132. #	.rva	function_unwind_info

  1133. #

  1134. # Reference to functon_unwind_info from .xdata segment is the anchor.

  1135. # In case you wonder why references are 32-bit .rvas and not 64-bit

  1136. # .quads. References put into these two segments are required to be

  1137. # *relative* to the base address of the current binary module, a.k.a.

  1138. # image base. No Win64 module, be it .exe or .dll, can be larger than

  1139. # 2GB and thus such relative references can be and are accommodated in

  1140. # 32 bits.

  1141. #

  1142. # Having reviewed the example function code, one can argue that "movq

  1143. # %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix

  1144. # rax would contain an undefined value. If this "offends" you, use

  1145. # another register and refrain from modifying rax till magic_point is

  1146. # reached, i.e. as if it was a non-volatile register. If more registers

  1147. # are required prior [variable] frame setup is completed, note that

  1148. # nobody says that you can have only one "magic point." You can

  1149. # "liberate" non-volatile registers by denoting last stack off-load

  1150. # instruction and reflecting it in finer grade unwind logic in handler.

  1151. # After all, isn't it why it's called *language-specific* handler...

  1152. #

  1153. # Attentive reader can notice that exceptions would be mishandled in

  1154. # auto-generated "gear" epilogue. Well, exception effectively can't

  1155. # occur there, because if memory area used by it was subject to

  1156. # segmentation violation, then it would be raised upon call to the

  1157. # function (and as already mentioned be accounted to caller, which is

  1158. # not a problem). If you're still not comfortable, then define tail

  1159. # "magic point" just prior ret instruction and have handler treat it...

  1160. #

  1161. # (*)	Note that we're talking about run-time, not debug-time. Lack of

  1162. #	unwind information makes debugging hard on both Windows and

  1163. #	Unix. "Unlike" referes to the fact that on Unix signal handler

  1164. #	will always be invoked, core dumped and appropriate exit code

  1165. #	returned to parent (for user notification).