1#! /usr/bin/env perl
2# Copyright 2005-2020 The OpenSSL Project Authors. All Rights Reserved.
3#
4# Licensed under the Apache License 2.0 (the "License").  You may not use
5# this file except in compliance with the License.  You can obtain a copy
6# in the file LICENSE in the source distribution or at
7# https://www.openssl.org/source/license.html
8
9
10# Ascetic x86_64 AT&T to MASM/NASM assembler translator by <appro>.
11#
12# Why AT&T to MASM and not vice versa? Several reasons. Because AT&T
13# format is way easier to parse. Because it's simpler to "gear" from
14# Unix ABI to Windows one [see cross-reference "card" at the end of
15# file]. Because Linux targets were available first...
16#
17# In addition the script also "distills" code suitable for GNU
18# assembler, so that it can be compiled with more rigid assemblers,
19# such as Solaris /usr/ccs/bin/as.
20#
21# This translator is not designed to convert *arbitrary* assembler
22# code from AT&T format to MASM one. It's designed to convert just
23# enough to provide for dual-ABI OpenSSL modules development...
24# There *are* limitations and you might have to modify your assembler
25# code or this script to achieve the desired result...
26#
27# Currently recognized limitations:
28#
29# - can't use multiple ops per line;
30#
31# Dual-ABI styling rules.
32#
33# 1. Adhere to Unix register and stack layout [see cross-reference
34#    ABI "card" at the end for explanation].
35# 2. Forget about "red zone," stick to more traditional blended
36#    stack frame allocation. If volatile storage is actually required
37#    that is. If not, just leave the stack as is.
38# 3. Functions tagged with ".type name,@function" get crafted with
39#    unified Win64 prologue and epilogue automatically. If you want
40#    to take care of ABI differences yourself, tag functions as
41#    ".type name,@abi-omnipotent" instead.
42# 4. To optimize the Win64 prologue you can specify number of input
43#    arguments as ".type name,@function,N." Keep in mind that if N is
44#    larger than 6, then you *have to* write "abi-omnipotent" code,
45#    because >6 cases can't be addressed with unified prologue.
46# 5. Name local labels as .L*, do *not* use dynamic labels such as 1:
47#    (sorry about latter).
48# 6. Don't use [or hand-code with .byte] "rep ret." "ret" mnemonic is
49#    required to identify the spots, where to inject Win64 epilogue!
50#    But on the pros, it's then prefixed with rep automatically:-)
51# 7. Stick to explicit ip-relative addressing. If you have to use
52#    GOTPCREL addressing, stick to mov symbol@GOTPCREL(%rip),%r??.
53#    Both are recognized and translated to proper Win64 addressing
54#    modes.
55#
56# 8. In order to provide for structured exception handling unified
57#    Win64 prologue copies %rsp value to %rax. For further details
58#    see SEH paragraph at the end.
59# 9. .init segment is allowed to contain calls to functions only.
60# a. If function accepts more than 4 arguments *and* >4th argument
61#    is declared as non 64-bit value, do clear its upper part.
62
63
64use strict;
65
66my $flavour = shift;
67my $output  = shift;
68if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
69
70open STDOUT,">$output" || die "can't open $output: $!"
71	if (defined($output));
72
73my $gas=1;	$gas=0 if ($output =~ /\.asm$/);
74my $elf=1;	$elf=0 if (!$gas);
75my $win64=0;
76my $prefix="";
77my $decor=".L";
78
79my $masmref=8 + 50727*2**-32;	# 8.00.50727 shipped with VS2005
80my $masm=0;
81my $PTR=" PTR";
82
83my $nasmref=2.03;
84my $nasm=0;
85
86# GNU as indicator, as opposed to $gas, which indicates acceptable
87# syntax
88my $gnuas=0;
89
90if    ($flavour eq "mingw64")	{ $gas=1; $elf=0; $win64=1;
91				  $prefix=`echo __USER_LABEL_PREFIX__ | $ENV{CC} -E -P -`;
92				  $prefix =~ s|\R$||; # Better chomp
93				}
94elsif ($flavour eq "macosx")	{ $gas=1; $elf=0; $prefix="_"; $decor="L\$"; }
95elsif ($flavour eq "masm")	{ $gas=0; $elf=0; $masm=$masmref; $win64=1; $decor="\$L\$"; }
96elsif ($flavour eq "nasm")	{ $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $decor="\$L\$"; $PTR=""; }
97elsif (!$gas)
98{   if ($ENV{ASM} =~ m/nasm/ && `nasm -v` =~ m/version ([0-9]+)\.([0-9]+)/i)
99    {	$nasm = $1 + $2*0.01; $PTR="";  }
100    elsif (`ml64 2>&1` =~ m/Version ([0-9]+)\.([0-9]+)(\.([0-9]+))?/)
101    {	$masm = $1 + $2*2**-16 + $4*2**-32;   }
102    die "no assembler found on %PATH%" if (!($nasm || $masm));
103    $win64=1;
104    $elf=0;
105    $decor="\$L\$";
106}
107# Find out if we're using GNU as
108elsif (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
109		=~ /GNU assembler version ([2-9]\.[0-9]+)/)
110{
111    $gnuas=1;
112}
113elsif (`$ENV{CC} --version 2>/dev/null`
114		=~ /(clang .*|Intel.*oneAPI .*)/)
115{
116    $gnuas=1;
117}
118elsif (`$ENV{CC} -V 2>/dev/null`
119		=~ /nvc .*/)
120{
121    $gnuas=1;
122}
123
124my $cet_property;
125if ($flavour =~ /elf/) {
126	# Always generate .note.gnu.property section for ELF outputs to
127	# mark Intel CET support since all input files must be marked
128	# with Intel CET support in order for linker to mark output with
129	# Intel CET support.
130	my $p2align=3; $p2align=2 if ($flavour eq "elf32");
131	my $section='.note.gnu.property, #alloc';
132	$section='".note.gnu.property", "a"' if $gnuas;
133	$cet_property = <<_____;
134	.section $section
135	.p2align $p2align
136	.long 1f - 0f
137	.long 4f - 1f
138	.long 5
1390:
140	# "GNU" encoded with .byte, since .asciz isn't supported
141	# on Solaris.
142	.byte 0x47
143	.byte 0x4e
144	.byte 0x55
145	.byte 0
1461:
147	.p2align $p2align
148	.long 0xc0000002
149	.long 3f - 2f
1502:
151	.long 3
1523:
153	.p2align $p2align
1544:
155_____
156}
157
158my $current_segment;
159my $current_function;
160my %globals;
161
162{ package opcode;	# pick up opcodes
163    sub re {
164	my	($class, $line) = @_;
165	my	$self = {};
166	my	$ret;
167
168	if ($$line =~ /^([a-z][a-z0-9]*)/i) {
169	    bless $self,$class;
170	    $self->{op} = $1;
171	    $ret = $self;
172	    $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
173
174	    undef $self->{sz};
175	    if ($self->{op} =~ /^(movz)x?([bw]).*/) {	# movz is pain...
176		$self->{op} = $1;
177		$self->{sz} = $2;
178	    } elsif ($self->{op} =~ /call|jmp/) {
179		$self->{sz} = "";
180	    } elsif ($self->{op} =~ /^p/ && $' !~ /^(ush|op|insrw)/) { # SSEn
181		$self->{sz} = "";
182	    } elsif ($self->{op} =~ /^[vk]/) { # VEX or k* such as kmov
183		$self->{sz} = "";
184	    } elsif ($self->{op} =~ /mov[dq]/ && $$line =~ /%xmm/) {
185		$self->{sz} = "";
186	    } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) {
187		$self->{op} = $1;
188		$self->{sz} = $2;
189	    }
190	}
191	$ret;
192    }
193    sub size {
194	my ($self, $sz) = @_;
195	$self->{sz} = $sz if (defined($sz) && !defined($self->{sz}));
196	$self->{sz};
197    }
198    sub out {
199	my $self = shift;
200	if ($gas) {
201	    if ($self->{op} eq "movz") {	# movz is pain...
202		sprintf "%s%s%s",$self->{op},$self->{sz},shift;
203	    } elsif ($self->{op} =~ /^set/) {
204		"$self->{op}";
205	    } elsif ($self->{op} eq "ret") {
206		my $epilogue = "";
207		if ($win64 && $current_function->{abi} eq "svr4") {
208		    $epilogue = "movq	8(%rsp),%rdi\n\t" .
209				"movq	16(%rsp),%rsi\n\t";
210		}
211	    	$epilogue . ".byte	0xf3,0xc3";
212	    } elsif ($self->{op} eq "call" && !$elf && $current_segment eq ".init") {
213		".p2align\t3\n\t.quad";
214	    } else {
215		"$self->{op}$self->{sz}";
216	    }
217	} else {
218	    $self->{op} =~ s/^movz/movzx/;
219	    if ($self->{op} eq "ret") {
220		$self->{op} = "";
221		if ($win64 && $current_function->{abi} eq "svr4") {
222		    $self->{op} = "mov	rdi,QWORD$PTR\[8+rsp\]\t;WIN64 epilogue\n\t".
223				  "mov	rsi,QWORD$PTR\[16+rsp\]\n\t";
224	    	}
225		$self->{op} .= "DB\t0F3h,0C3h\t\t;repret";
226	    } elsif ($self->{op} =~ /^(pop|push)f/) {
227		$self->{op} .= $self->{sz};
228	    } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") {
229		$self->{op} = "\tDQ";
230	    }
231	    $self->{op};
232	}
233    }
234    sub mnemonic {
235	my ($self, $op) = @_;
236	$self->{op}=$op if (defined($op));
237	$self->{op};
238    }
239}
240{ package const;	# pick up constants, which start with $
241    sub re {
242	my	($class, $line) = @_;
243	my	$self = {};
244	my	$ret;
245
246	if ($$line =~ /^\$([^,]+)/) {
247	    bless $self, $class;
248	    $self->{value} = $1;
249	    $ret = $self;
250	    $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
251	}
252	$ret;
253    }
254    sub out {
255    	my $self = shift;
256
257	$self->{value} =~ s/\b(0b[0-1]+)/oct($1)/eig;
258	if ($gas) {
259	    # Solaris /usr/ccs/bin/as can't handle multiplications
260	    # in $self->{value}
261	    my $value = $self->{value};
262	    no warnings;    # oct might complain about overflow, ignore here...
263	    $value =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;
264	    if ($value =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg) {
265		$self->{value} = $value;
266	    }
267	    sprintf "\$%s",$self->{value};
268	} else {
269	    my $value = $self->{value};
270	    $value =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm);
271	    sprintf "%s",$value;
272	}
273    }
274}
275{ package ea;		# pick up effective addresses: expr(%reg,%reg,scale)
276
277    my %szmap = (	b=>"BYTE$PTR",    w=>"WORD$PTR",
278			l=>"DWORD$PTR",   d=>"DWORD$PTR",
279			q=>"QWORD$PTR",   o=>"OWORD$PTR",
280			x=>"XMMWORD$PTR", y=>"YMMWORD$PTR",
281			z=>"ZMMWORD$PTR" ) if (!$gas);
282
283    sub re {
284	my	($class, $line, $opcode) = @_;
285	my	$self = {};
286	my	$ret;
287
288	# optional * ----vvv--- appears in indirect jmp/call
289	if ($$line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)((?:{[^}]+})*)/) {
290	    bless $self, $class;
291	    $self->{asterisk} = $1;
292	    $self->{label} = $2;
293	    ($self->{base},$self->{index},$self->{scale})=split(/,/,$3);
294	    $self->{scale} = 1 if (!defined($self->{scale}));
295	    $self->{opmask} = $4;
296	    $ret = $self;
297	    $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
298
299	    if ($win64 && $self->{label} =~ s/\@GOTPCREL//) {
300		die if ($opcode->mnemonic() ne "mov");
301		$opcode->mnemonic("lea");
302	    }
303	    $self->{base}  =~ s/^%//;
304	    $self->{index} =~ s/^%// if (defined($self->{index}));
305	    $self->{opcode} = $opcode;
306	}
307	$ret;
308    }
309    sub size {}
310    sub out {
311	my ($self, $sz) = @_;
312
313	$self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
314	$self->{label} =~ s/\.L/$decor/g;
315
316	# Silently convert all EAs to 64-bit. This is required for
317	# elder GNU assembler and results in more compact code,
318	# *but* most importantly AES module depends on this feature!
319	$self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
320	$self->{base}  =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
321
322	# Solaris /usr/ccs/bin/as can't handle multiplications
323	# in $self->{label}...
324	use integer;
325	$self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;
326	$self->{label} =~ s/\b([0-9]+\s*[\*\/\%]\s*[0-9]+)\b/eval($1)/eg;
327
328	# Some assemblers insist on signed presentation of 32-bit
329	# offsets, but sign extension is a tricky business in perl...
330	if ((1<<31)<<1) {
331	    $self->{label} =~ s/\b([0-9]+)\b/$1<<32>>32/eg;
332	} else {
333	    $self->{label} =~ s/\b([0-9]+)\b/$1>>0/eg;
334	}
335
336	# if base register is %rbp or %r13, see if it's possible to
337	# flip base and index registers [for better performance]
338	if (!$self->{label} && $self->{index} && $self->{scale}==1 &&
339	    $self->{base} =~ /(rbp|r13)/) {
340		$self->{base} = $self->{index}; $self->{index} = $1;
341	}
342
343	if ($gas) {
344	    $self->{label} =~ s/^___imp_/__imp__/   if ($flavour eq "mingw64");
345
346	    if (defined($self->{index})) {
347		sprintf "%s%s(%s,%%%s,%d)%s",
348					$self->{asterisk},$self->{label},
349					$self->{base}?"%$self->{base}":"",
350					$self->{index},$self->{scale},
351					$self->{opmask};
352	    } else {
353		sprintf "%s%s(%%%s)%s",	$self->{asterisk},$self->{label},
354					$self->{base},$self->{opmask};
355	    }
356	} else {
357	    $self->{label} =~ s/\./\$/g;
358	    $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig;
359	    $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/);
360
361	    my $mnemonic = $self->{opcode}->mnemonic();
362	    ($self->{asterisk})				&& ($sz="q") ||
363	    ($mnemonic =~ /^v?mov([qd])$/)		&& ($sz=$1)  ||
364	    ($mnemonic =~ /^v?pinsr([qdwb])$/)		&& ($sz=$1)  ||
365	    ($mnemonic =~ /^vpbroadcast([qdwb])$/)	&& ($sz=$1)  ||
366	    ($mnemonic =~ /^v(?!perm)[a-z]+[fi]128$/)	&& ($sz="x");
367
368	    $self->{opmask}  =~ s/%(k[0-7])/$1/;
369
370	    if (defined($self->{index})) {
371		sprintf "%s[%s%s*%d%s]%s",$szmap{$sz},
372					$self->{label}?"$self->{label}+":"",
373					$self->{index},$self->{scale},
374					$self->{base}?"+$self->{base}":"",
375					$self->{opmask};
376	    } elsif ($self->{base} eq "rip") {
377		sprintf "%s[%s]",$szmap{$sz},$self->{label};
378	    } else {
379		sprintf "%s[%s%s]%s",	$szmap{$sz},
380					$self->{label}?"$self->{label}+":"",
381					$self->{base},$self->{opmask};
382	    }
383	}
384    }
385}
386{ package register;	# pick up registers, which start with %.
387    sub re {
388	my	($class, $line, $opcode) = @_;
389	my	$self = {};
390	my	$ret;
391
392	# optional * ----vvv--- appears in indirect jmp/call
393	if ($$line =~ /^(\*?)%(\w+)((?:{[^}]+})*)/) {
394	    bless $self,$class;
395	    $self->{asterisk} = $1;
396	    $self->{value} = $2;
397	    $self->{opmask} = $3;
398	    $opcode->size($self->size());
399	    $ret = $self;
400	    $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
401	}
402	$ret;
403    }
404    sub size {
405	my	$self = shift;
406	my	$ret;
407
408	if    ($self->{value} =~ /^r[\d]+b$/i)	{ $ret="b"; }
409	elsif ($self->{value} =~ /^r[\d]+w$/i)	{ $ret="w"; }
410	elsif ($self->{value} =~ /^r[\d]+d$/i)	{ $ret="l"; }
411	elsif ($self->{value} =~ /^r[\w]+$/i)	{ $ret="q"; }
412	elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; }
413	elsif ($self->{value} =~ /^[\w]{2}l$/i)	{ $ret="b"; }
414	elsif ($self->{value} =~ /^[\w]{2}$/i)	{ $ret="w"; }
415	elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; }
416
417	$ret;
418    }
419    sub out {
420    	my $self = shift;
421	if ($gas)	{ sprintf "%s%%%s%s",	$self->{asterisk},
422						$self->{value},
423						$self->{opmask}; }
424	else		{ $self->{opmask} =~ s/%(k[0-7])/$1/;
425			  $self->{value}.$self->{opmask}; }
426    }
427}
428{ package label;	# pick up labels, which end with :
429    sub re {
430	my	($class, $line) = @_;
431	my	$self = {};
432	my	$ret;
433
434	if ($$line =~ /(^[\.\w]+)\:/) {
435	    bless $self,$class;
436	    $self->{value} = $1;
437	    $ret = $self;
438	    $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
439
440	    $self->{value} =~ s/^\.L/$decor/;
441	}
442	$ret;
443    }
444    sub out {
445	my $self = shift;
446
447	if ($gas) {
448	    my $func = ($globals{$self->{value}} or $self->{value}) . ":";
449	    if ($win64	&& $current_function->{name} eq $self->{value}
450			&& $current_function->{abi} eq "svr4") {
451		$func .= "\n";
452		$func .= "	movq	%rdi,8(%rsp)\n";
453		$func .= "	movq	%rsi,16(%rsp)\n";
454		$func .= "	movq	%rsp,%rax\n";
455		$func .= "${decor}SEH_begin_$current_function->{name}:\n";
456		my $narg = $current_function->{narg};
457		$narg=6 if (!defined($narg));
458		$func .= "	movq	%rcx,%rdi\n" if ($narg>0);
459		$func .= "	movq	%rdx,%rsi\n" if ($narg>1);
460		$func .= "	movq	%r8,%rdx\n"  if ($narg>2);
461		$func .= "	movq	%r9,%rcx\n"  if ($narg>3);
462		$func .= "	movq	40(%rsp),%r8\n" if ($narg>4);
463		$func .= "	movq	48(%rsp),%r9\n" if ($narg>5);
464	    }
465	    $func;
466	} elsif ($self->{value} ne "$current_function->{name}") {
467	    # Make all labels in masm global.
468	    $self->{value} .= ":" if ($masm);
469	    $self->{value} . ":";
470	} elsif ($win64 && $current_function->{abi} eq "svr4") {
471	    my $func =	"$current_function->{name}" .
472			($nasm ? ":" : "\tPROC $current_function->{scope}") .
473			"\n";
474	    $func .= "	mov	QWORD$PTR\[8+rsp\],rdi\t;WIN64 prologue\n";
475	    $func .= "	mov	QWORD$PTR\[16+rsp\],rsi\n";
476	    $func .= "	mov	rax,rsp\n";
477	    $func .= "${decor}SEH_begin_$current_function->{name}:";
478	    $func .= ":" if ($masm);
479	    $func .= "\n";
480	    my $narg = $current_function->{narg};
481	    $narg=6 if (!defined($narg));
482	    $func .= "	mov	rdi,rcx\n" if ($narg>0);
483	    $func .= "	mov	rsi,rdx\n" if ($narg>1);
484	    $func .= "	mov	rdx,r8\n"  if ($narg>2);
485	    $func .= "	mov	rcx,r9\n"  if ($narg>3);
486	    $func .= "	mov	r8,QWORD$PTR\[40+rsp\]\n" if ($narg>4);
487	    $func .= "	mov	r9,QWORD$PTR\[48+rsp\]\n" if ($narg>5);
488	    $func .= "\n";
489	} else {
490	   "$current_function->{name}".
491			($nasm ? ":" : "\tPROC $current_function->{scope}");
492	}
493    }
494}
495{ package expr;		# pick up expressions
496    sub re {
497	my	($class, $line, $opcode) = @_;
498	my	$self = {};
499	my	$ret;
500
501	if ($$line =~ /(^[^,]+)/) {
502	    bless $self,$class;
503	    $self->{value} = $1;
504	    $ret = $self;
505	    $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
506
507	    $self->{value} =~ s/\@PLT// if (!$elf);
508	    $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
509	    $self->{value} =~ s/\.L/$decor/g;
510	    $self->{opcode} = $opcode;
511	}
512	$ret;
513    }
514    sub out {
515	my $self = shift;
516	if ($nasm && $self->{opcode}->mnemonic()=~m/^j(?![re]cxz)/) {
517	    "NEAR ".$self->{value};
518	} else {
519	    $self->{value};
520	}
521    }
522}
523{ package cfi_directive;
524    # CFI directives annotate instructions that are significant for
525    # stack unwinding procedure compliant with DWARF specification,
526    # see http://dwarfstd.org/. Besides naturally expected for this
527    # script platform-specific filtering function, this module adds
528    # three auxiliary synthetic directives not recognized by [GNU]
529    # assembler:
530    #
531    # - .cfi_push to annotate push instructions in prologue, which
532    #   translates to .cfi_adjust_cfa_offset (if needed) and
533    #   .cfi_offset;
534    # - .cfi_pop to annotate pop instructions in epilogue, which
535    #   translates to .cfi_adjust_cfa_offset (if needed) and
536    #   .cfi_restore;
537    # - [and most notably] .cfi_cfa_expression which encodes
538    #   DW_CFA_def_cfa_expression and passes it to .cfi_escape as
539    #   byte vector;
540    #
541    # CFA expressions were introduced in DWARF specification version
542    # 3 and describe how to deduce CFA, Canonical Frame Address. This
543    # becomes handy if your stack frame is variable and you can't
544    # spare register for [previous] frame pointer. Suggested directive
545    # syntax is made-up mix of DWARF operator suffixes [subset of]
546    # and references to registers with optional bias. Following example
547    # describes offloaded *original* stack pointer at specific offset
548    # from *current* stack pointer:
549    #
550    #   .cfi_cfa_expression     %rsp+40,deref,+8
551    #
552    # Final +8 has everything to do with the fact that CFA is defined
553    # as reference to top of caller's stack, and on x86_64 call to
554    # subroutine pushes 8-byte return address. In other words original
555    # stack pointer upon entry to a subroutine is 8 bytes off from CFA.
556
557    # Below constants are taken from "DWARF Expressions" section of the
558    # DWARF specification, section is numbered 7.7 in versions 3 and 4.
559    my %DW_OP_simple = (	# no-arg operators, mapped directly
560	deref	=> 0x06,	dup	=> 0x12,
561	drop	=> 0x13,	over	=> 0x14,
562	pick	=> 0x15,	swap	=> 0x16,
563	rot	=> 0x17,	xderef	=> 0x18,
564
565	abs	=> 0x19,	and	=> 0x1a,
566	div	=> 0x1b,	minus	=> 0x1c,
567	mod	=> 0x1d,	mul	=> 0x1e,
568	neg	=> 0x1f,	not	=> 0x20,
569	or	=> 0x21,	plus	=> 0x22,
570	shl	=> 0x24,	shr	=> 0x25,
571	shra	=> 0x26,	xor	=> 0x27,
572	);
573
574    my %DW_OP_complex = (	# used in specific subroutines
575	constu		=> 0x10,	# uleb128
576	consts		=> 0x11,	# sleb128
577	plus_uconst	=> 0x23,	# uleb128
578	lit0 		=> 0x30,	# add 0-31 to opcode
579	reg0		=> 0x50,	# add 0-31 to opcode
580	breg0		=> 0x70,	# add 0-31 to opcole, sleb128
581	regx		=> 0x90,	# uleb28
582	fbreg		=> 0x91,	# sleb128
583	bregx		=> 0x92,	# uleb128, sleb128
584	piece		=> 0x93,	# uleb128
585	);
586
587    # Following constants are defined in x86_64 ABI supplement, for
588    # example available at https://www.uclibc.org/docs/psABI-x86_64.pdf,
589    # see section 3.7 "Stack Unwind Algorithm".
590    my %DW_reg_idx = (
591	"%rax"=>0,  "%rdx"=>1,  "%rcx"=>2,  "%rbx"=>3,
592	"%rsi"=>4,  "%rdi"=>5,  "%rbp"=>6,  "%rsp"=>7,
593	"%r8" =>8,  "%r9" =>9,  "%r10"=>10, "%r11"=>11,
594	"%r12"=>12, "%r13"=>13, "%r14"=>14, "%r15"=>15
595	);
596
597    my ($cfa_reg, $cfa_rsp);
598    my @cfa_stack;
599
600    # [us]leb128 format is variable-length integer representation base
601    # 2^128, with most significant bit of each byte being 0 denoting
602    # *last* most significant digit. See "Variable Length Data" in the
603    # DWARF specification, numbered 7.6 at least in versions 3 and 4.
604    sub sleb128 {
605	use integer;	# get right shift extend sign
606
607	my $val = shift;
608	my $sign = ($val < 0) ? -1 : 0;
609	my @ret = ();
610
611	while(1) {
612	    push @ret, $val&0x7f;
613
614	    # see if remaining bits are same and equal to most
615	    # significant bit of the current digit, if so, it's
616	    # last digit...
617	    last if (($val>>6) == $sign);
618
619	    @ret[-1] |= 0x80;
620	    $val >>= 7;
621	}
622
623	return @ret;
624    }
625    sub uleb128 {
626	my $val = shift;
627	my @ret = ();
628
629	while(1) {
630	    push @ret, $val&0x7f;
631
632	    # see if it's last significant digit...
633	    last if (($val >>= 7) == 0);
634
635	    @ret[-1] |= 0x80;
636	}
637
638	return @ret;
639    }
640    sub const {
641	my $val = shift;
642
643	if ($val >= 0 && $val < 32) {
644            return ($DW_OP_complex{lit0}+$val);
645	}
646	return ($DW_OP_complex{consts}, sleb128($val));
647    }
648    sub reg {
649	my $val = shift;
650
651	return if ($val !~ m/^(%r\w+)(?:([\+\-])((?:0x)?[0-9a-f]+))?/);
652
653	my $reg = $DW_reg_idx{$1};
654	my $off = eval ("0 $2 $3");
655
656	return (($DW_OP_complex{breg0} + $reg), sleb128($off));
657	# Yes, we use DW_OP_bregX+0 to push register value and not
658	# DW_OP_regX, because latter would require even DW_OP_piece,
659	# which would be a waste under the circumstances. If you have
660	# to use DWP_OP_reg, use "regx:N"...
661    }
662    sub cfa_expression {
663	my $line = shift;
664	my @ret;
665
666	foreach my $token (split(/,\s*/,$line)) {
667	    if ($token =~ /^%r/) {
668		push @ret,reg($token);
669	    } elsif ($token =~ /((?:0x)?[0-9a-f]+)\((%r\w+)\)/) {
670		push @ret,reg("$2+$1");
671	    } elsif ($token =~ /(\w+):(\-?(?:0x)?[0-9a-f]+)(U?)/i) {
672		my $i = 1*eval($2);
673		push @ret,$DW_OP_complex{$1}, ($3 ? uleb128($i) : sleb128($i));
674	    } elsif (my $i = 1*eval($token) or $token eq "0") {
675		if ($token =~ /^\+/) {
676		    push @ret,$DW_OP_complex{plus_uconst},uleb128($i);
677		} else {
678		    push @ret,const($i);
679		}
680	    } else {
681		push @ret,$DW_OP_simple{$token};
682	    }
683	}
684
685	# Finally we return DW_CFA_def_cfa_expression, 15, followed by
686	# length of the expression and of course the expression itself.
687	return (15,scalar(@ret),@ret);
688    }
689    sub re {
690	my	($class, $line) = @_;
691	my	$self = {};
692	my	$ret;
693
694	if ($$line =~ s/^\s*\.cfi_(\w+)\s*//) {
695	    bless $self,$class;
696	    $ret = $self;
697	    undef $self->{value};
698	    my $dir = $1;
699
700	    SWITCH: for ($dir) {
701	    # What is $cfa_rsp? Effectively it's difference between %rsp
702	    # value and current CFA, Canonical Frame Address, which is
703	    # why it starts with -8. Recall that CFA is top of caller's
704	    # stack...
705	    /startproc/	&& do {	($cfa_reg, $cfa_rsp) = ("%rsp", -8); last; };
706	    /endproc/	&& do {	($cfa_reg, $cfa_rsp) = ("%rsp",  0);
707				# .cfi_remember_state directives that are not
708				# matched with .cfi_restore_state are
709				# unnecessary.
710				die "unpaired .cfi_remember_state" if (@cfa_stack);
711				last;
712			      };
713	    /def_cfa_register/
714			&& do {	$cfa_reg = $$line; last; };
715	    /def_cfa_offset/
716			&& do {	$cfa_rsp = -1*eval($$line) if ($cfa_reg eq "%rsp");
717				last;
718			      };
719	    /adjust_cfa_offset/
720			&& do {	$cfa_rsp -= 1*eval($$line) if ($cfa_reg eq "%rsp");
721				last;
722			      };
723	    /def_cfa/	&& do {	if ($$line =~ /(%r\w+)\s*,\s*(.+)/) {
724				    $cfa_reg = $1;
725				    $cfa_rsp = -1*eval($2) if ($cfa_reg eq "%rsp");
726				}
727				last;
728			      };
729	    /push/	&& do {	$dir = undef;
730				$cfa_rsp -= 8;
731				if ($cfa_reg eq "%rsp") {
732				    $self->{value} = ".cfi_adjust_cfa_offset\t8\n";
733				}
734				$self->{value} .= ".cfi_offset\t$$line,$cfa_rsp";
735				last;
736			      };
737	    /pop/	&& do {	$dir = undef;
738				$cfa_rsp += 8;
739				if ($cfa_reg eq "%rsp") {
740				    $self->{value} = ".cfi_adjust_cfa_offset\t-8\n";
741				}
742				$self->{value} .= ".cfi_restore\t$$line";
743				last;
744			      };
745	    /cfa_expression/
746			&& do {	$dir = undef;
747				$self->{value} = ".cfi_escape\t" .
748					join(",", map(sprintf("0x%02x", $_),
749						      cfa_expression($$line)));
750				last;
751			      };
752	    /remember_state/
753			&& do {	push @cfa_stack, [$cfa_reg, $cfa_rsp];
754				last;
755			      };
756	    /restore_state/
757			&& do {	($cfa_reg, $cfa_rsp) = @{pop @cfa_stack};
758				last;
759			      };
760	    }
761
762	    $self->{value} = ".cfi_$dir\t$$line" if ($dir);
763
764	    $$line = "";
765	}
766
767	return $ret;
768    }
769    sub out {
770	my $self = shift;
771	return ($elf ? $self->{value} : undef);
772    }
773}
774{ package directive;	# pick up directives, which start with .
775    sub re {
776	my	($class, $line) = @_;
777	my	$self = {};
778	my	$ret;
779	my	$dir;
780
781	# chain-call to cfi_directive
782	$ret = cfi_directive->re($line) and return $ret;
783
784	if ($$line =~ /^\s*(\.\w+)/) {
785	    bless $self,$class;
786	    $dir = $1;
787	    $ret = $self;
788	    undef $self->{value};
789	    $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
790
791	    SWITCH: for ($dir) {
792		/\.global|\.globl|\.extern/
793			    && do { $globals{$$line} = $prefix . $$line;
794				    $$line = $globals{$$line} if ($prefix);
795				    last;
796				  };
797		/\.type/    && do { my ($sym,$type,$narg) = split(',',$$line);
798				    if ($type eq "\@function") {
799					undef $current_function;
800					$current_function->{name} = $sym;
801					$current_function->{abi}  = "svr4";
802					$current_function->{narg} = $narg;
803					$current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";
804				    } elsif ($type eq "\@abi-omnipotent") {
805					undef $current_function;
806					$current_function->{name} = $sym;
807					$current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";
808				    }
809				    $$line =~ s/\@abi\-omnipotent/\@function/;
810				    $$line =~ s/\@function.*/\@function/;
811				    last;
812				  };
813		/\.asciz/   && do { if ($$line =~ /^"(.*)"$/) {
814					$dir  = ".byte";
815					$$line = join(",",unpack("C*",$1),0);
816				    }
817				    last;
818				  };
819		/\.rva|\.long|\.quad/
820			    && do { $$line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
821				    $$line =~ s/\.L/$decor/g;
822				    last;
823				  };
824	    }
825
826	    if ($gas) {
827		$self->{value} = $dir . "\t" . $$line;
828
829		if ($dir =~ /\.extern/) {
830		    $self->{value} = ""; # swallow extern
831		} elsif (!$elf && $dir =~ /\.type/) {
832		    $self->{value} = "";
833		    $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" .
834				(defined($globals{$1})?".scl 2;":".scl 3;") .
835				"\t.type 32;\t.endef"
836				if ($win64 && $$line =~ /([^,]+),\@function/);
837		} elsif (!$elf && $dir =~ /\.size/) {
838		    $self->{value} = "";
839		    if (defined($current_function)) {
840			$self->{value} .= "${decor}SEH_end_$current_function->{name}:"
841				if ($win64 && $current_function->{abi} eq "svr4");
842			undef $current_function;
843		    }
844		} elsif (!$elf && $dir =~ /\.align/) {
845		    $self->{value} = ".p2align\t" . (log($$line)/log(2));
846		} elsif ($dir eq ".section") {
847		    $current_segment=$$line;
848		    if (!$elf && $current_segment eq ".init") {
849			if	($flavour eq "macosx")	{ $self->{value} = ".mod_init_func"; }
850			elsif	($flavour eq "mingw64")	{ $self->{value} = ".section\t.ctors"; }
851		    }
852		} elsif ($dir =~ /\.(text|data)/) {
853		    $current_segment=".$1";
854		} elsif ($dir =~ /\.hidden/) {
855		    if    ($flavour eq "macosx")  { $self->{value} = ".private_extern\t$prefix$$line"; }
856		    elsif ($flavour eq "mingw64") { $self->{value} = ""; }
857		} elsif ($dir =~ /\.comm/) {
858		    $self->{value} = "$dir\t$prefix$$line";
859		    $self->{value} =~ s|,([0-9]+),([0-9]+)$|",$1,".log($2)/log(2)|e if ($flavour eq "macosx");
860		}
861		$$line = "";
862		return $self;
863	    }
864
865	    # non-gas case or nasm/masm
866	    SWITCH: for ($dir) {
867		/\.text/    && do { my $v=undef;
868				    if ($nasm) {
869					$v="section	.text code align=64\n";
870				    } else {
871					$v="$current_segment\tENDS\n" if ($current_segment);
872					$current_segment = ".text\$";
873					$v.="$current_segment\tSEGMENT ";
874					$v.=$masm>=$masmref ? "ALIGN(256)" : "PAGE";
875					$v.=" 'CODE'";
876				    }
877				    $self->{value} = $v;
878				    last;
879				  };
880		/\.data/    && do { my $v=undef;
881				    if ($nasm) {
882					$v="section	.data data align=8\n";
883				    } else {
884					$v="$current_segment\tENDS\n" if ($current_segment);
885					$current_segment = "_DATA";
886					$v.="$current_segment\tSEGMENT";
887				    }
888				    $self->{value} = $v;
889				    last;
890				  };
891		/\.section/ && do { my $v=undef;
892				    $$line =~ s/([^,]*).*/$1/;
893				    $$line = ".CRT\$XCU" if ($$line eq ".init");
894				    if ($nasm) {
895					$v="section	$$line";
896					if ($$line=~/\.([px])data/) {
897					    $v.=" rdata align=";
898					    $v.=$1 eq "p"? 4 : 8;
899					} elsif ($$line=~/\.CRT\$/i) {
900					    $v.=" rdata align=8";
901					}
902				    } else {
903					$v="$current_segment\tENDS\n" if ($current_segment);
904					$v.="$$line\tSEGMENT";
905					if ($$line=~/\.([px])data/) {
906					    $v.=" READONLY";
907					    $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref);
908					} elsif ($$line=~/\.CRT\$/i) {
909					    $v.=" READONLY ";
910					    $v.=$masm>=$masmref ? "ALIGN(8)" : "DWORD";
911					}
912				    }
913				    $current_segment = $$line;
914				    $self->{value} = $v;
915				    last;
916				  };
917		/\.extern/  && do { $self->{value}  = "EXTERN\t".$$line;
918				    $self->{value} .= ":NEAR" if ($masm);
919				    last;
920				  };
921		/\.globl|.global/
922			    && do { $self->{value}  = $masm?"PUBLIC":"global";
923				    $self->{value} .= "\t".$$line;
924				    last;
925				  };
926		/\.size/    && do { if (defined($current_function)) {
927					undef $self->{value};
928					if ($current_function->{abi} eq "svr4") {
929					    $self->{value}="${decor}SEH_end_$current_function->{name}:";
930					    $self->{value}.=":\n" if($masm);
931					}
932					$self->{value}.="$current_function->{name}\tENDP" if($masm && $current_function->{name});
933					undef $current_function;
934				    }
935				    last;
936				  };
937		/\.align/   && do { my $max = ($masm && $masm>=$masmref) ? 256 : 4096;
938				    $self->{value} = "ALIGN\t".($$line>$max?$max:$$line);
939				    last;
940				  };
941		/\.(value|long|rva|quad)/
942			    && do { my $sz  = substr($1,0,1);
943				    my @arr = split(/,\s*/,$$line);
944				    my $last = pop(@arr);
945				    my $conv = sub  {	my $var=shift;
946							$var=~s/^(0b[0-1]+)/oct($1)/eig;
947							$var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm);
948							if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva"))
949							{ $var=~s/^([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; }
950							$var;
951						    };
952
953				    $sz =~ tr/bvlrq/BWDDQ/;
954				    $self->{value} = "\tD$sz\t";
955				    for (@arr) { $self->{value} .= &$conv($_).","; }
956				    $self->{value} .= &$conv($last);
957				    last;
958				  };
959		/\.byte/    && do { my @str=split(/,\s*/,$$line);
960				    map(s/(0b[0-1]+)/oct($1)/eig,@str);
961				    map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm);
962				    while ($#str>15) {
963					$self->{value}.="DB\t"
964						.join(",",@str[0..15])."\n";
965					foreach (0..15) { shift @str; }
966				    }
967				    $self->{value}.="DB\t"
968						.join(",",@str) if (@str);
969				    last;
970				  };
971		/\.comm/    && do { my @str=split(/,\s*/,$$line);
972				    my $v=undef;
973				    if ($nasm) {
974					$v.="common	$prefix@str[0] @str[1]";
975				    } else {
976					$v="$current_segment\tENDS\n" if ($current_segment);
977					$current_segment = "_DATA";
978					$v.="$current_segment\tSEGMENT\n";
979					$v.="COMM	@str[0]:DWORD:".@str[1]/4;
980				    }
981				    $self->{value} = $v;
982				    last;
983				  };
984	    }
985	    $$line = "";
986	}
987
988	$ret;
989    }
990    sub out {
991	my $self = shift;
992	$self->{value};
993    }
994}
995
996# Upon initial x86_64 introduction SSE>2 extensions were not introduced
997# yet. In order not to be bothered by tracing exact assembler versions,
998# but at the same time to provide a bare security minimum of AES-NI, we
999# hard-code some instructions. Extensions past AES-NI on the other hand
1000# are traced by examining assembler version in individual perlasm
1001# modules...
1002
1003my %regrm = (	"%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3,
1004		"%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7	);
1005
1006sub rex {
1007 my $opcode=shift;
1008 my ($dst,$src,$rex)=@_;
1009
1010   $rex|=0x04 if($dst>=8);
1011   $rex|=0x01 if($src>=8);
1012   push @$opcode,($rex|0x40) if ($rex);
1013}
1014
1015my $movq = sub {	# elderly gas can't handle inter-register movq
1016  my $arg = shift;
1017  my @opcode=(0x66);
1018    if ($arg =~ /%xmm([0-9]+),\s*%r(\w+)/) {
1019	my ($src,$dst)=($1,$2);
1020	if ($dst !~ /[0-9]+/)	{ $dst = $regrm{"%e$dst"}; }
1021	rex(\@opcode,$src,$dst,0x8);
1022	push @opcode,0x0f,0x7e;
1023	push @opcode,0xc0|(($src&7)<<3)|($dst&7);	# ModR/M
1024	@opcode;
1025    } elsif ($arg =~ /%r(\w+),\s*%xmm([0-9]+)/) {
1026	my ($src,$dst)=($2,$1);
1027	if ($dst !~ /[0-9]+/)	{ $dst = $regrm{"%e$dst"}; }
1028	rex(\@opcode,$src,$dst,0x8);
1029	push @opcode,0x0f,0x6e;
1030	push @opcode,0xc0|(($src&7)<<3)|($dst&7);	# ModR/M
1031	@opcode;
1032    } else {
1033	();
1034    }
1035};
1036
1037my $pextrd = sub {
1038    if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*(%\w+)/) {
1039      my @opcode=(0x66);
1040	my $imm=$1;
1041	my $src=$2;
1042	my $dst=$3;
1043	if ($dst =~ /%r([0-9]+)d/)	{ $dst = $1; }
1044	elsif ($dst =~ /%e/)		{ $dst = $regrm{$dst}; }
1045	rex(\@opcode,$src,$dst);
1046	push @opcode,0x0f,0x3a,0x16;
1047	push @opcode,0xc0|(($src&7)<<3)|($dst&7);	# ModR/M
1048	push @opcode,$imm;
1049	@opcode;
1050    } else {
1051	();
1052    }
1053};
1054
1055my $pinsrd = sub {
1056    if (shift =~ /\$([0-9]+),\s*(%\w+),\s*%xmm([0-9]+)/) {
1057      my @opcode=(0x66);
1058	my $imm=$1;
1059	my $src=$2;
1060	my $dst=$3;
1061	if ($src =~ /%r([0-9]+)/)	{ $src = $1; }
1062	elsif ($src =~ /%e/)		{ $src = $regrm{$src}; }
1063	rex(\@opcode,$dst,$src);
1064	push @opcode,0x0f,0x3a,0x22;
1065	push @opcode,0xc0|(($dst&7)<<3)|($src&7);	# ModR/M
1066	push @opcode,$imm;
1067	@opcode;
1068    } else {
1069	();
1070    }
1071};
1072
1073my $pshufb = sub {
1074    if (shift =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1075      my @opcode=(0x66);
1076	rex(\@opcode,$2,$1);
1077	push @opcode,0x0f,0x38,0x00;
1078	push @opcode,0xc0|($1&7)|(($2&7)<<3);		# ModR/M
1079	@opcode;
1080    } else {
1081	();
1082    }
1083};
1084
1085my $palignr = sub {
1086    if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1087      my @opcode=(0x66);
1088	rex(\@opcode,$3,$2);
1089	push @opcode,0x0f,0x3a,0x0f;
1090	push @opcode,0xc0|($2&7)|(($3&7)<<3);		# ModR/M
1091	push @opcode,$1;
1092	@opcode;
1093    } else {
1094	();
1095    }
1096};
1097
1098my $pclmulqdq = sub {
1099    if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1100      my @opcode=(0x66);
1101	rex(\@opcode,$3,$2);
1102	push @opcode,0x0f,0x3a,0x44;
1103	push @opcode,0xc0|($2&7)|(($3&7)<<3);		# ModR/M
1104	my $c=$1;
1105	push @opcode,$c=~/^0/?oct($c):$c;
1106	@opcode;
1107    } else {
1108	();
1109    }
1110};
1111
1112my $rdrand = sub {
1113    if (shift =~ /%[er](\w+)/) {
1114      my @opcode=();
1115      my $dst=$1;
1116	if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
1117	rex(\@opcode,0,$dst,8);
1118	push @opcode,0x0f,0xc7,0xf0|($dst&7);
1119	@opcode;
1120    } else {
1121	();
1122    }
1123};
1124
1125my $rdseed = sub {
1126    if (shift =~ /%[er](\w+)/) {
1127      my @opcode=();
1128      my $dst=$1;
1129	if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
1130	rex(\@opcode,0,$dst,8);
1131	push @opcode,0x0f,0xc7,0xf8|($dst&7);
1132	@opcode;
1133    } else {
1134	();
1135    }
1136};
1137
1138# Not all AVX-capable assemblers recognize AMD XOP extension. Since we
1139# are using only two instructions hand-code them in order to be excused
1140# from chasing assembler versions...
1141
1142sub rxb {
1143 my $opcode=shift;
1144 my ($dst,$src1,$src2,$rxb)=@_;
1145
1146   $rxb|=0x7<<5;
1147   $rxb&=~(0x04<<5) if($dst>=8);
1148   $rxb&=~(0x01<<5) if($src1>=8);
1149   $rxb&=~(0x02<<5) if($src2>=8);
1150   push @$opcode,$rxb;
1151}
1152
1153my $vprotd = sub {
1154    if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1155      my @opcode=(0x8f);
1156	rxb(\@opcode,$3,$2,-1,0x08);
1157	push @opcode,0x78,0xc2;
1158	push @opcode,0xc0|($2&7)|(($3&7)<<3);		# ModR/M
1159	my $c=$1;
1160	push @opcode,$c=~/^0/?oct($c):$c;
1161	@opcode;
1162    } else {
1163	();
1164    }
1165};
1166
1167my $vprotq = sub {
1168    if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1169      my @opcode=(0x8f);
1170	rxb(\@opcode,$3,$2,-1,0x08);
1171	push @opcode,0x78,0xc3;
1172	push @opcode,0xc0|($2&7)|(($3&7)<<3);		# ModR/M
1173	my $c=$1;
1174	push @opcode,$c=~/^0/?oct($c):$c;
1175	@opcode;
1176    } else {
1177	();
1178    }
1179};
1180
1181# Intel Control-flow Enforcement Technology extension. All functions and
1182# indirect branch targets will have to start with this instruction...
1183
1184my $endbranch = sub {
1185    (0xf3,0x0f,0x1e,0xfa);
1186};
1187
1188########################################################################
1189
1190if ($nasm) {
1191    print <<___;
1192default	rel
1193%define XMMWORD
1194%define YMMWORD
1195%define ZMMWORD
1196___
1197} elsif ($masm) {
1198    print <<___;
1199OPTION	DOTNAME
1200___
1201}
1202while(defined(my $line=<>)) {
1203
1204    $line =~ s|\R$||;           # Better chomp
1205
1206    $line =~ s|[#!].*$||;	# get rid of asm-style comments...
1207    $line =~ s|/\*.*\*/||;	# ... and C-style comments...
1208    $line =~ s|^\s+||;		# ... and skip whitespaces in beginning
1209    $line =~ s|\s+$||;		# ... and at the end
1210
1211    if (my $label=label->re(\$line))	{ print $label->out(); }
1212
1213    if (my $directive=directive->re(\$line)) {
1214	printf "%s",$directive->out();
1215    } elsif (my $opcode=opcode->re(\$line)) {
1216	my $asm = eval("\$".$opcode->mnemonic());
1217
1218	if ((ref($asm) eq 'CODE') && scalar(my @bytes=&$asm($line))) {
1219	    print $gas?".byte\t":"DB\t",join(',',@bytes),"\n";
1220	    next;
1221	}
1222
1223	my @args;
1224	ARGUMENT: while (1) {
1225	    my $arg;
1226
1227	    ($arg=register->re(\$line, $opcode))||
1228	    ($arg=const->re(\$line))		||
1229	    ($arg=ea->re(\$line, $opcode))	||
1230	    ($arg=expr->re(\$line, $opcode))	||
1231	    last ARGUMENT;
1232
1233	    push @args,$arg;
1234
1235	    last ARGUMENT if ($line !~ /^,/);
1236
1237	    $line =~ s/^,\s*//;
1238	} # ARGUMENT:
1239
1240	if ($#args>=0) {
1241	    my $insn;
1242	    my $sz=$opcode->size();
1243
1244	    if ($gas) {
1245		$insn = $opcode->out($#args>=1?$args[$#args]->size():$sz);
1246		@args = map($_->out($sz),@args);
1247		printf "\t%s\t%s",$insn,join(",",@args);
1248	    } else {
1249		$insn = $opcode->out();
1250		foreach (@args) {
1251		    my $arg = $_->out();
1252		    # $insn.=$sz compensates for movq, pinsrw, ...
1253		    if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; }
1254		    if ($arg =~ /^ymm[0-9]+$/) { $insn.=$sz; $sz="y" if(!$sz); last; }
1255		    if ($arg =~ /^zmm[0-9]+$/) { $insn.=$sz; $sz="z" if(!$sz); last; }
1256		    if ($arg =~ /^mm[0-9]+$/)  { $insn.=$sz; $sz="q" if(!$sz); last; }
1257		}
1258		@args = reverse(@args);
1259		undef $sz if ($nasm && $opcode->mnemonic() eq "lea");
1260		printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args));
1261	    }
1262	} else {
1263	    printf "\t%s",$opcode->out();
1264	}
1265    }
1266
1267    print $line,"\n";
1268}
1269
1270print "$cet_property"			if ($cet_property);
1271print "\n$current_segment\tENDS\n"	if ($current_segment && $masm);
1272print "END\n"				if ($masm);
1273
1274close STDOUT or die "error closing STDOUT: $!;"
1275
1276#################################################
1277# Cross-reference x86_64 ABI "card"
1278#
1279# 		Unix		Win64
1280# %rax		*		*
1281# %rbx		-		-
1282# %rcx		#4		#1
1283# %rdx		#3		#2
1284# %rsi		#2		-
1285# %rdi		#1		-
1286# %rbp		-		-
1287# %rsp		-		-
1288# %r8		#5		#3
1289# %r9		#6		#4
1290# %r10		*		*
1291# %r11		*		*
1292# %r12		-		-
1293# %r13		-		-
1294# %r14		-		-
1295# %r15		-		-
1296#
1297# (*)	volatile register
1298# (-)	preserved by callee
1299# (#)	Nth argument, volatile
1300#
1301# In Unix terms top of stack is argument transfer area for arguments
1302# which could not be accommodated in registers. Or in other words 7th
1303# [integer] argument resides at 8(%rsp) upon function entry point.
1304# 128 bytes above %rsp constitute a "red zone" which is not touched
1305# by signal handlers and can be used as temporal storage without
1306# allocating a frame.
1307#
1308# In Win64 terms N*8 bytes on top of stack is argument transfer area,
1309# which belongs to/can be overwritten by callee. N is the number of
1310# arguments passed to callee, *but* not less than 4! This means that
1311# upon function entry point 5th argument resides at 40(%rsp), as well
1312# as that 32 bytes from 8(%rsp) can always be used as temporal
1313# storage [without allocating a frame]. One can actually argue that
1314# one can assume a "red zone" above stack pointer under Win64 as well.
1315# Point is that at apparently no occasion Windows kernel would alter
1316# the area above user stack pointer in true asynchronous manner...
1317#
1318# All the above means that if assembler programmer adheres to Unix
1319# register and stack layout, but disregards the "red zone" existence,
1320# it's possible to use following prologue and epilogue to "gear" from
1321# Unix to Win64 ABI in leaf functions with not more than 6 arguments.
1322#
1323# omnipotent_function:
1324# ifdef WIN64
1325#	movq	%rdi,8(%rsp)
1326#	movq	%rsi,16(%rsp)
1327#	movq	%rcx,%rdi	; if 1st argument is actually present
1328#	movq	%rdx,%rsi	; if 2nd argument is actually ...
1329#	movq	%r8,%rdx	; if 3rd argument is ...
1330#	movq	%r9,%rcx	; if 4th argument ...
1331#	movq	40(%rsp),%r8	; if 5th ...
1332#	movq	48(%rsp),%r9	; if 6th ...
1333# endif
1334#	...
1335# ifdef WIN64
1336#	movq	8(%rsp),%rdi
1337#	movq	16(%rsp),%rsi
1338# endif
1339#	ret
1340#
1341#################################################
1342# Win64 SEH, Structured Exception Handling.
1343#
1344# Unlike on Unix systems(*) lack of Win64 stack unwinding information
1345# has undesired side-effect at run-time: if an exception is raised in
1346# assembler subroutine such as those in question (basically we're
1347# referring to segmentation violations caused by malformed input
1348# parameters), the application is briskly terminated without invoking
1349# any exception handlers, most notably without generating memory dump
1350# or any user notification whatsoever. This poses a problem. It's
1351# possible to address it by registering custom language-specific
1352# handler that would restore processor context to the state at
1353# subroutine entry point and return "exception is not handled, keep
1354# unwinding" code. Writing such handler can be a challenge... But it's
1355# doable, though requires certain coding convention. Consider following
1356# snippet:
1357#
1358# .type	function,@function
1359# function:
1360#	movq	%rsp,%rax	# copy rsp to volatile register
1361#	pushq	%r15		# save non-volatile registers
1362#	pushq	%rbx
1363#	pushq	%rbp
1364#	movq	%rsp,%r11
1365#	subq	%rdi,%r11	# prepare [variable] stack frame
1366#	andq	$-64,%r11
1367#	movq	%rax,0(%r11)	# check for exceptions
1368#	movq	%r11,%rsp	# allocate [variable] stack frame
1369#	movq	%rax,0(%rsp)	# save original rsp value
1370# magic_point:
1371#	...
1372#	movq	0(%rsp),%rcx	# pull original rsp value
1373#	movq	-24(%rcx),%rbp	# restore non-volatile registers
1374#	movq	-16(%rcx),%rbx
1375#	movq	-8(%rcx),%r15
1376#	movq	%rcx,%rsp	# restore original rsp
1377# magic_epilogue:
1378#	ret
1379# .size function,.-function
1380#
1381# The key is that up to magic_point copy of original rsp value remains
1382# in chosen volatile register and no non-volatile register, except for
1383# rsp, is modified. While past magic_point rsp remains constant till
1384# the very end of the function. In this case custom language-specific
1385# exception handler would look like this:
1386#
1387# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1388#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
1389# {	ULONG64 *rsp = (ULONG64 *)context->Rax;
1390#	ULONG64  rip = context->Rip;
1391#
1392#	if (rip >= magic_point)
1393#	{   rsp = (ULONG64 *)context->Rsp;
1394#	    if (rip < magic_epilogue)
1395#	    {	rsp = (ULONG64 *)rsp[0];
1396#		context->Rbp = rsp[-3];
1397#		context->Rbx = rsp[-2];
1398#		context->R15 = rsp[-1];
1399#	    }
1400#	}
1401#	context->Rsp = (ULONG64)rsp;
1402#	context->Rdi = rsp[1];
1403#	context->Rsi = rsp[2];
1404#
1405#	memcpy (disp->ContextRecord,context,sizeof(CONTEXT));
1406#	RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase,
1407#		dips->ControlPc,disp->FunctionEntry,disp->ContextRecord,
1408#		&disp->HandlerData,&disp->EstablisherFrame,NULL);
1409#	return ExceptionContinueSearch;
1410# }
1411#
1412# It's appropriate to implement this handler in assembler, directly in
1413# function's module. In order to do that one has to know members'
1414# offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant
1415# values. Here they are:
1416#
1417#	CONTEXT.Rax				120
1418#	CONTEXT.Rcx				128
1419#	CONTEXT.Rdx				136
1420#	CONTEXT.Rbx				144
1421#	CONTEXT.Rsp				152
1422#	CONTEXT.Rbp				160
1423#	CONTEXT.Rsi				168
1424#	CONTEXT.Rdi				176
1425#	CONTEXT.R8				184
1426#	CONTEXT.R9				192
1427#	CONTEXT.R10				200
1428#	CONTEXT.R11				208
1429#	CONTEXT.R12				216
1430#	CONTEXT.R13				224
1431#	CONTEXT.R14				232
1432#	CONTEXT.R15				240
1433#	CONTEXT.Rip				248
1434#	CONTEXT.Xmm6				512
1435#	sizeof(CONTEXT)				1232
1436#	DISPATCHER_CONTEXT.ControlPc		0
1437#	DISPATCHER_CONTEXT.ImageBase		8
1438#	DISPATCHER_CONTEXT.FunctionEntry	16
1439#	DISPATCHER_CONTEXT.EstablisherFrame	24
1440#	DISPATCHER_CONTEXT.TargetIp		32
1441#	DISPATCHER_CONTEXT.ContextRecord	40
1442#	DISPATCHER_CONTEXT.LanguageHandler	48
1443#	DISPATCHER_CONTEXT.HandlerData		56
1444#	UNW_FLAG_NHANDLER			0
1445#	ExceptionContinueSearch			1
1446#
1447# In order to tie the handler to the function one has to compose
1448# couple of structures: one for .xdata segment and one for .pdata.
1449#
1450# UNWIND_INFO structure for .xdata segment would be
1451#
1452# function_unwind_info:
1453#	.byte	9,0,0,0
1454#	.rva	handler
1455#
1456# This structure designates exception handler for a function with
1457# zero-length prologue, no stack frame or frame register.
1458#
1459# To facilitate composing of .pdata structures, auto-generated "gear"
1460# prologue copies rsp value to rax and denotes next instruction with
1461# .LSEH_begin_{function_name} label. This essentially defines the SEH
1462# styling rule mentioned in the beginning. Position of this label is
1463# chosen in such manner that possible exceptions raised in the "gear"
1464# prologue would be accounted to caller and unwound from latter's frame.
1465# End of function is marked with respective .LSEH_end_{function_name}
1466# label. To summarize, .pdata segment would contain
1467#
1468#	.rva	.LSEH_begin_function
1469#	.rva	.LSEH_end_function
1470#	.rva	function_unwind_info
1471#
1472# Reference to function_unwind_info from .xdata segment is the anchor.
1473# In case you wonder why references are 32-bit .rvas and not 64-bit
1474# .quads. References put into these two segments are required to be
1475# *relative* to the base address of the current binary module, a.k.a.
1476# image base. No Win64 module, be it .exe or .dll, can be larger than
1477# 2GB and thus such relative references can be and are accommodated in
1478# 32 bits.
1479#
1480# Having reviewed the example function code, one can argue that "movq
1481# %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix
1482# rax would contain an undefined value. If this "offends" you, use
1483# another register and refrain from modifying rax till magic_point is
1484# reached, i.e. as if it was a non-volatile register. If more registers
1485# are required prior [variable] frame setup is completed, note that
1486# nobody says that you can have only one "magic point." You can
1487# "liberate" non-volatile registers by denoting last stack off-load
1488# instruction and reflecting it in finer grade unwind logic in handler.
1489# After all, isn't it why it's called *language-specific* handler...
1490#
1491# SE handlers are also involved in unwinding stack when executable is
1492# profiled or debugged. Profiling implies additional limitations that
1493# are too subtle to discuss here. For now it's sufficient to say that
1494# in order to simplify handlers one should either a) offload original
1495# %rsp to stack (like discussed above); or b) if you have a register to
1496# spare for frame pointer, choose volatile one.
1497#
1498# (*)	Note that we're talking about run-time, not debug-time. Lack of
1499#	unwind information makes debugging hard on both Windows and
1500#	Unix. "Unlike" refers to the fact that on Unix signal handler
1501#	will always be invoked, core dumped and appropriate exit code
1502#	returned to parent (for user notification).
1503