2 # Copyright 2005-2016 The OpenSSL Project Authors. All Rights Reserved.
4 # Licensed under the OpenSSL license (the "License"). You may not use
5 # this file except in compliance with the License. You can obtain a copy
6 # in the file LICENSE in the source distribution or at
7 # https://www.openssl.org/source/license.html
10 # Ascetic x86_64 AT&T to MASM/NASM assembler translator by <appro>.
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...
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.
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...
27 # Currently recognized limitations:
29 # - can't use multiple ops per line;
31 # Dual-ABI styling rules.
33 # 1. Adhere to Unix register and stack layout [see cross-reference
34 # ABI "card" at the end for explanation].
35 # 2. Forget about "red zone," stick to more traditional blended
36 # stack frame allocation. If volatile storage is actually required
37 # that is. If not, just leave the stack as is.
38 # 3. Functions tagged with ".type name,@function" get crafted with
39 # unified Win64 prologue and epilogue automatically. If you want
40 # to take care of ABI differences yourself, tag functions as
41 # ".type name,@abi-omnipotent" instead.
42 # 4. To optimize the Win64 prologue you can specify number of input
43 # arguments as ".type name,@function,N." Keep in mind that if N is
44 # larger than 6, then you *have to* write "abi-omnipotent" code,
45 # because >6 cases can't be addressed with unified prologue.
46 # 5. Name local labels as .L*, do *not* use dynamic labels such as 1:
47 # (sorry about latter).
48 # 6. Don't use [or hand-code with .byte] "rep ret." "ret" mnemonic is
49 # required to identify the spots, where to inject Win64 epilogue!
50 # But on the pros, it's then prefixed with rep automatically:-)
51 # 7. Stick to explicit ip-relative addressing. If you have to use
52 # GOTPCREL addressing, stick to mov symbol@GOTPCREL(%rip),%r??.
53 # Both are recognized and translated to proper Win64 addressing
54 # modes. To support legacy code a synthetic directive, .picmeup,
55 # is implemented. It puts address of the *next* instruction into
56 # target register, e.g.:
59 # lea .Label-.(%rax),%rax
61 # 8. In order to provide for structured exception handling unified
62 # Win64 prologue copies %rsp value to %rax. For further details
63 # see SEH paragraph at the end.
64 # 9. .init segment is allowed to contain calls to functions only.
65 # a. If function accepts more than 4 arguments *and* >4th argument
66 # is declared as non 64-bit value, do clear its upper part.
73 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
75 open STDOUT,">$output" || die "can't open $output: $!"
76 if (defined($output));
78 my $gas=1; $gas=0 if ($output =~ /\.asm$/);
79 my $elf=1; $elf=0 if (!$gas);
84 my $masmref=8 + 50727*2**-32; # 8.00.50727 shipped with VS2005
91 if ($flavour eq "mingw64") { $gas=1; $elf=0; $win64=1;
92 $prefix=`echo __USER_LABEL_PREFIX__ | $ENV{CC} -E -P -`;
93 $prefix =~ s|\R$||; # Better chomp
95 elsif ($flavour eq "macosx") { $gas=1; $elf=0; $prefix="_"; $decor="L\$"; }
96 elsif ($flavour eq "masm") { $gas=0; $elf=0; $masm=$masmref; $win64=1; $decor="\$L\$"; }
97 elsif ($flavour eq "nasm") { $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $decor="\$L\$"; $PTR=""; }
99 { if ($ENV{ASM} =~ m/nasm/ && `nasm -v` =~ m/version ([0-9]+)\.([0-9]+)/i)
100 { $nasm = $1 + $2*0.01; $PTR=""; }
101 elsif (`ml64 2>&1` =~ m/Version ([0-9]+)\.([0-9]+)(\.([0-9]+))?/)
102 { $masm = $1 + $2*2**-16 + $4*2**-32; }
103 die "no assembler found on %PATH" if (!($nasm || $masm));
110 my $current_function;
113 { package opcode; # pick up opcodes
115 my ($class, $line) = @_;
119 if ($$line =~ /^([a-z][a-z0-9]*)/i) {
123 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
126 if ($self->{op} =~ /^(movz)x?([bw]).*/) { # movz is pain...
129 } elsif ($self->{op} =~ /call|jmp/) {
131 } elsif ($self->{op} =~ /^p/ && $' !~ /^(ush|op|insrw)/) { # SSEn
133 } elsif ($self->{op} =~ /^[vk]/) { # VEX or k* such as kmov
135 } elsif ($self->{op} =~ /mov[dq]/ && $$line =~ /%xmm/) {
137 } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) {
145 my ($self, $sz) = @_;
146 $self->{sz} = $sz if (defined($sz) && !defined($self->{sz}));
152 if ($self->{op} eq "movz") { # movz is pain...
153 sprintf "%s%s%s",$self->{op},$self->{sz},shift;
154 } elsif ($self->{op} =~ /^set/) {
156 } elsif ($self->{op} eq "ret") {
158 if ($win64 && $current_function->{abi} eq "svr4") {
159 $epilogue = "movq 8(%rsp),%rdi\n\t" .
160 "movq 16(%rsp),%rsi\n\t";
162 $epilogue . ".byte 0xf3,0xc3";
163 } elsif ($self->{op} eq "call" && !$elf && $current_segment eq ".init") {
164 ".p2align\t3\n\t.quad";
166 "$self->{op}$self->{sz}";
169 $self->{op} =~ s/^movz/movzx/;
170 if ($self->{op} eq "ret") {
172 if ($win64 && $current_function->{abi} eq "svr4") {
173 $self->{op} = "mov rdi,QWORD$PTR\[8+rsp\]\t;WIN64 epilogue\n\t".
174 "mov rsi,QWORD$PTR\[16+rsp\]\n\t";
176 $self->{op} .= "DB\t0F3h,0C3h\t\t;repret";
177 } elsif ($self->{op} =~ /^(pop|push)f/) {
178 $self->{op} .= $self->{sz};
179 } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") {
180 $self->{op} = "\tDQ";
186 my ($self, $op) = @_;
187 $self->{op}=$op if (defined($op));
191 { package const; # pick up constants, which start with $
193 my ($class, $line) = @_;
197 if ($$line =~ /^\$([^,]+)/) {
201 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
208 $self->{value} =~ s/\b(0b[0-1]+)/oct($1)/eig;
210 # Solaris /usr/ccs/bin/as can't handle multiplications
212 my $value = $self->{value};
213 no warnings; # oct might complain about overflow, ignore here...
214 $value =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;
215 if ($value =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg) {
216 $self->{value} = $value;
218 sprintf "\$%s",$self->{value};
220 $self->{value} =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm);
221 sprintf "%s",$self->{value};
225 { package ea; # pick up effective addresses: expr(%reg,%reg,scale)
227 my ($class, $line, $opcode) = @_;
231 # optional * ----vvv--- appears in indirect jmp/call
232 if ($$line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)((?:{[^}]+})*)/) {
234 $self->{asterisk} = $1;
236 ($self->{base},$self->{index},$self->{scale})=split(/,/,$3);
237 $self->{scale} = 1 if (!defined($self->{scale}));
240 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
242 if ($win64 && $self->{label} =~ s/\@GOTPCREL//) {
243 die if ($opcode->mnemonic() ne "mov");
244 $opcode->mnemonic("lea");
246 $self->{base} =~ s/^%//;
247 $self->{index} =~ s/^%// if (defined($self->{index}));
248 $self->{opcode} = $opcode;
254 my ($self, $sz) = @_;
256 $self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
257 $self->{label} =~ s/\.L/$decor/g;
259 # Silently convert all EAs to 64-bit. This is required for
260 # elder GNU assembler and results in more compact code,
261 # *but* most importantly AES module depends on this feature!
262 $self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
263 $self->{base} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
265 # Solaris /usr/ccs/bin/as can't handle multiplications
266 # in $self->{label}...
268 $self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;
269 $self->{label} =~ s/\b([0-9]+\s*[\*\/\%]\s*[0-9]+)\b/eval($1)/eg;
271 # Some assemblers insist on signed presentation of 32-bit
272 # offsets, but sign extension is a tricky business in perl...
274 $self->{label} =~ s/\b([0-9]+)\b/$1<<32>>32/eg;
276 $self->{label} =~ s/\b([0-9]+)\b/$1>>0/eg;
279 if (!$self->{label} && $self->{index} && $self->{scale}==1 &&
280 $self->{base} =~ /(rbp|r13)/) {
281 $self->{base} = $self->{index}; $self->{index} = $1;
285 $self->{label} =~ s/^___imp_/__imp__/ if ($flavour eq "mingw64");
287 if (defined($self->{index})) {
288 sprintf "%s%s(%s,%%%s,%d)%s",
289 $self->{asterisk},$self->{label},
290 $self->{base}?"%$self->{base}":"",
291 $self->{index},$self->{scale},
294 sprintf "%s%s(%%%s)%s", $self->{asterisk},$self->{label},
295 $self->{base},$self->{pred};
298 my %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR",
299 l=>"DWORD$PTR", d=>"DWORD$PTR",
300 q=>"QWORD$PTR", o=>"OWORD$PTR",
301 x=>"XMMWORD$PTR", y=>"YMMWORD$PTR", z=>"ZMMWORD$PTR" );
303 $self->{label} =~ s/\./\$/g;
304 $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig;
305 $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/);
307 my $mnemonic = $self->{opcode}->mnemonic();
308 ($self->{asterisk}) && ($sz="q") ||
309 ($mnemonic =~ /^v?mov([qd])$/) && ($sz=$1) ||
310 ($mnemonic =~ /^v?pinsr([qdwb])$/) && ($sz=$1) ||
311 ($mnemonic =~ /^vpbroadcast([qdwb])$/) && ($sz=$1) ||
312 ($mnemonic =~ /^v(?!perm)[a-z]+[fi]128$/) && ($sz="x");
314 $self->{pred} =~ s/%(k[0-7])/$1/;
316 if (defined($self->{index})) {
317 sprintf "%s[%s%s*%d%s]%s",$szmap{$sz},
318 $self->{label}?"$self->{label}+":"",
319 $self->{index},$self->{scale},
320 $self->{base}?"+$self->{base}":"",
322 } elsif ($self->{base} eq "rip") {
323 sprintf "%s[%s]",$szmap{$sz},$self->{label};
325 sprintf "%s[%s%s]%s", $szmap{$sz},
326 $self->{label}?"$self->{label}+":"",
327 $self->{base},$self->{pred};
332 { package register; # pick up registers, which start with %.
334 my ($class, $line, $opcode) = @_;
338 # optional * ----vvv--- appears in indirect jmp/call
339 if ($$line =~ /^(\*?)%(\w+)((?:{[^}]+})*)/) {
341 $self->{asterisk} = $1;
344 $opcode->size($self->size());
346 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
354 if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; }
355 elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; }
356 elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; }
357 elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; }
358 elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; }
359 elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; }
360 elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; }
361 elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; }
367 if ($gas) { sprintf "%s%%%s%s", $self->{asterisk},
370 else { $self->{pred} =~ s/%(k[0-7])/$1/;
371 $self->{value}.$self->{pred}; }
374 { package label; # pick up labels, which end with :
376 my ($class, $line) = @_;
380 if ($$line =~ /(^[\.\w]+)\:/) {
384 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
386 $self->{value} =~ s/^\.L/$decor/;
394 my $func = ($globals{$self->{value}} or $self->{value}) . ":";
396 $current_function->{name} eq $self->{value} &&
397 $current_function->{abi} eq "svr4") {
399 $func .= " movq %rdi,8(%rsp)\n";
400 $func .= " movq %rsi,16(%rsp)\n";
401 $func .= " movq %rsp,%rax\n";
402 $func .= "${decor}SEH_begin_$current_function->{name}:\n";
403 my $narg = $current_function->{narg};
404 $narg=6 if (!defined($narg));
405 $func .= " movq %rcx,%rdi\n" if ($narg>0);
406 $func .= " movq %rdx,%rsi\n" if ($narg>1);
407 $func .= " movq %r8,%rdx\n" if ($narg>2);
408 $func .= " movq %r9,%rcx\n" if ($narg>3);
409 $func .= " movq 40(%rsp),%r8\n" if ($narg>4);
410 $func .= " movq 48(%rsp),%r9\n" if ($narg>5);
413 } elsif ($self->{value} ne "$current_function->{name}") {
414 # Make all labels in masm global.
415 $self->{value} .= ":" if ($masm);
416 $self->{value} . ":";
417 } elsif ($win64 && $current_function->{abi} eq "svr4") {
418 my $func = "$current_function->{name}" .
419 ($nasm ? ":" : "\tPROC $current_function->{scope}") .
421 $func .= " mov QWORD$PTR\[8+rsp\],rdi\t;WIN64 prologue\n";
422 $func .= " mov QWORD$PTR\[16+rsp\],rsi\n";
423 $func .= " mov rax,rsp\n";
424 $func .= "${decor}SEH_begin_$current_function->{name}:";
425 $func .= ":" if ($masm);
427 my $narg = $current_function->{narg};
428 $narg=6 if (!defined($narg));
429 $func .= " mov rdi,rcx\n" if ($narg>0);
430 $func .= " mov rsi,rdx\n" if ($narg>1);
431 $func .= " mov rdx,r8\n" if ($narg>2);
432 $func .= " mov rcx,r9\n" if ($narg>3);
433 $func .= " mov r8,QWORD$PTR\[40+rsp\]\n" if ($narg>4);
434 $func .= " mov r9,QWORD$PTR\[48+rsp\]\n" if ($narg>5);
437 "$current_function->{name}".
438 ($nasm ? ":" : "\tPROC $current_function->{scope}");
442 { package expr; # pick up expressioins
444 my ($class, $line, $opcode) = @_;
448 if ($$line =~ /(^[^,]+)/) {
452 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
454 $self->{value} =~ s/\@PLT// if (!$elf);
455 $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
456 $self->{value} =~ s/\.L/$decor/g;
457 $self->{opcode} = $opcode;
463 if ($nasm && $self->{opcode}->mnemonic()=~m/^j(?![re]cxz)/) {
464 "NEAR ".$self->{value};
470 { package directive; # pick up directives, which start with .
472 my ($class, $line) = @_;
476 my %opcode = # lea 2f-1f(%rip),%dst; 1: nop; 2:
477 ( "%rax"=>0x01058d48, "%rcx"=>0x010d8d48,
478 "%rdx"=>0x01158d48, "%rbx"=>0x011d8d48,
479 "%rsp"=>0x01258d48, "%rbp"=>0x012d8d48,
480 "%rsi"=>0x01358d48, "%rdi"=>0x013d8d48,
481 "%r8" =>0x01058d4c, "%r9" =>0x010d8d4c,
482 "%r10"=>0x01158d4c, "%r11"=>0x011d8d4c,
483 "%r12"=>0x01258d4c, "%r13"=>0x012d8d4c,
484 "%r14"=>0x01358d4c, "%r15"=>0x013d8d4c );
486 if ($$line =~ /^\s*(\.\w+)/) {
490 undef $self->{value};
491 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
494 /\.picmeup/ && do { if ($$line =~ /(%r[\w]+)/i) {
496 $$line=sprintf "0x%x,0x90000000",$opcode{$1};
500 /\.global|\.globl|\.extern/
501 && do { $globals{$$line} = $prefix . $$line;
502 $$line = $globals{$$line} if ($prefix);
505 /\.type/ && do { my ($sym,$type,$narg) = split(',',$$line);
506 if ($type eq "\@function") {
507 undef $current_function;
508 $current_function->{name} = $sym;
509 $current_function->{abi} = "svr4";
510 $current_function->{narg} = $narg;
511 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";
512 } elsif ($type eq "\@abi-omnipotent") {
513 undef $current_function;
514 $current_function->{name} = $sym;
515 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";
517 $$line =~ s/\@abi\-omnipotent/\@function/;
518 $$line =~ s/\@function.*/\@function/;
521 /\.asciz/ && do { if ($$line =~ /^"(.*)"$/) {
523 $$line = join(",",unpack("C*",$1),0);
527 /\.rva|\.long|\.quad/
528 && do { $$line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
529 $$line =~ s/\.L/$decor/g;
535 $self->{value} = $dir . "\t" . $$line;
537 if ($dir =~ /\.extern/) {
538 $self->{value} = ""; # swallow extern
539 } elsif (!$elf && $dir =~ /\.type/) {
541 $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" .
542 (defined($globals{$1})?".scl 2;":".scl 3;") .
543 "\t.type 32;\t.endef"
544 if ($win64 && $$line =~ /([^,]+),\@function/);
545 } elsif (!$elf && $dir =~ /\.size/) {
547 if (defined($current_function)) {
548 $self->{value} .= "${decor}SEH_end_$current_function->{name}:"
549 if ($win64 && $current_function->{abi} eq "svr4");
550 undef $current_function;
552 } elsif (!$elf && $dir =~ /\.align/) {
553 $self->{value} = ".p2align\t" . (log($$line)/log(2));
554 } elsif ($dir eq ".section") {
555 $current_segment=$$line;
556 if (!$elf && $current_segment eq ".init") {
557 if ($flavour eq "macosx") { $self->{value} = ".mod_init_func"; }
558 elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.ctors"; }
560 } elsif ($dir =~ /\.(text|data)/) {
561 $current_segment=".$1";
562 } elsif ($dir =~ /\.hidden/) {
563 if ($flavour eq "macosx") { $self->{value} = ".private_extern\t$prefix$$line"; }
564 elsif ($flavour eq "mingw64") { $self->{value} = ""; }
565 } elsif ($dir =~ /\.comm/) {
566 $self->{value} = "$dir\t$prefix$$line";
567 $self->{value} =~ s|,([0-9]+),([0-9]+)$|",$1,".log($2)/log(2)|e if ($flavour eq "macosx");
573 # non-gas case or nasm/masm
575 /\.text/ && do { my $v=undef;
577 $v="section .text code align=64\n";
579 $v="$current_segment\tENDS\n" if ($current_segment);
580 $current_segment = ".text\$";
581 $v.="$current_segment\tSEGMENT ";
582 $v.=$masm>=$masmref ? "ALIGN(256)" : "PAGE";
588 /\.data/ && do { my $v=undef;
590 $v="section .data data align=8\n";
592 $v="$current_segment\tENDS\n" if ($current_segment);
593 $current_segment = "_DATA";
594 $v.="$current_segment\tSEGMENT";
599 /\.section/ && do { my $v=undef;
600 $$line =~ s/([^,]*).*/$1/;
601 $$line = ".CRT\$XCU" if ($$line eq ".init");
604 if ($$line=~/\.([px])data/) {
606 $v.=$1 eq "p"? 4 : 8;
607 } elsif ($$line=~/\.CRT\$/i) {
608 $v.=" rdata align=8";
611 $v="$current_segment\tENDS\n" if ($current_segment);
612 $v.="$$line\tSEGMENT";
613 if ($$line=~/\.([px])data/) {
615 $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref);
616 } elsif ($$line=~/\.CRT\$/i) {
618 $v.=$masm>=$masmref ? "ALIGN(8)" : "DWORD";
621 $current_segment = $$line;
625 /\.extern/ && do { $self->{value} = "EXTERN\t".$$line;
626 $self->{value} .= ":NEAR" if ($masm);
630 && do { $self->{value} = $masm?"PUBLIC":"global";
631 $self->{value} .= "\t".$$line;
634 /\.size/ && do { if (defined($current_function)) {
635 undef $self->{value};
636 if ($current_function->{abi} eq "svr4") {
637 $self->{value}="${decor}SEH_end_$current_function->{name}:";
638 $self->{value}.=":\n" if($masm);
640 $self->{value}.="$current_function->{name}\tENDP" if($masm && $current_function->{name});
641 undef $current_function;
645 /\.align/ && do { my $max = ($masm && $masm>=$masmref) ? 256 : 4096;
646 $self->{value} = "ALIGN\t".($$line>$max?$max:$$line);
649 /\.(value|long|rva|quad)/
650 && do { my $sz = substr($1,0,1);
651 my @arr = split(/,\s*/,$$line);
652 my $last = pop(@arr);
653 my $conv = sub { my $var=shift;
654 $var=~s/^(0b[0-1]+)/oct($1)/eig;
655 $var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm);
656 if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva"))
657 { $var=~s/([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; }
661 $sz =~ tr/bvlrq/BWDDQ/;
662 $self->{value} = "\tD$sz\t";
663 for (@arr) { $self->{value} .= &$conv($_).","; }
664 $self->{value} .= &$conv($last);
667 /\.byte/ && do { my @str=split(/,\s*/,$$line);
668 map(s/(0b[0-1]+)/oct($1)/eig,@str);
669 map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm);
671 $self->{value}.="DB\t"
672 .join(",",@str[0..15])."\n";
673 foreach (0..15) { shift @str; }
675 $self->{value}.="DB\t"
676 .join(",",@str) if (@str);
679 /\.comm/ && do { my @str=split(/,\s*/,$$line);
682 $v.="common $prefix@str[0] @str[1]";
684 $v="$current_segment\tENDS\n" if ($current_segment);
685 $current_segment = "_DATA";
686 $v.="$current_segment\tSEGMENT\n";
687 $v.="COMM @str[0]:DWORD:".@str[1]/4;
706 my ($dst,$src,$rex)=@_;
708 $rex|=0x04 if($dst>=8);
709 $rex|=0x01 if($src>=8);
710 push @$opcode,($rex|0x40) if ($rex);
713 # Upon initial x86_64 introduction SSE>2 extensions were not introduced
714 # yet. In order not to be bothered by tracing exact assembler versions,
715 # but at the same time to provide a bare security minimum of AES-NI, we
716 # hard-code some instructions. Extensions past AES-NI on the other hand
717 # are traced by examining assembler version in individual perlasm
720 my %regrm = ( "%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3,
721 "%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7 );
723 my $movq = sub { # elderly gas can't handle inter-register movq
726 if ($arg =~ /%xmm([0-9]+),\s*%r(\w+)/) {
727 my ($src,$dst)=($1,$2);
728 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
729 rex(\@opcode,$src,$dst,0x8);
730 push @opcode,0x0f,0x7e;
731 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M
733 } elsif ($arg =~ /%r(\w+),\s*%xmm([0-9]+)/) {
734 my ($src,$dst)=($2,$1);
735 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
736 rex(\@opcode,$src,$dst,0x8);
737 push @opcode,0x0f,0x6e;
738 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M
746 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*(%\w+)/) {
751 if ($dst =~ /%r([0-9]+)d/) { $dst = $1; }
752 elsif ($dst =~ /%e/) { $dst = $regrm{$dst}; }
753 rex(\@opcode,$src,$dst);
754 push @opcode,0x0f,0x3a,0x16;
755 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M
764 if (shift =~ /\$([0-9]+),\s*(%\w+),\s*%xmm([0-9]+)/) {
769 if ($src =~ /%r([0-9]+)/) { $src = $1; }
770 elsif ($src =~ /%e/) { $src = $regrm{$src}; }
771 rex(\@opcode,$dst,$src);
772 push @opcode,0x0f,0x3a,0x22;
773 push @opcode,0xc0|(($dst&7)<<3)|($src&7); # ModR/M
782 if (shift =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) {
785 push @opcode,0x0f,0x38,0x00;
786 push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M
794 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
797 push @opcode,0x0f,0x3a,0x0f;
798 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
806 my $pclmulqdq = sub {
807 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
810 push @opcode,0x0f,0x3a,0x44;
811 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
813 push @opcode,$c=~/^0/?oct($c):$c;
821 if (shift =~ /%[er](\w+)/) {
824 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
825 rex(\@opcode,0,$dst,8);
826 push @opcode,0x0f,0xc7,0xf0|($dst&7);
834 if (shift =~ /%[er](\w+)/) {
837 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
838 rex(\@opcode,0,$dst,8);
839 push @opcode,0x0f,0xc7,0xf8|($dst&7);
848 my ($dst,$src1,$src2,$rxb)=@_;
851 $rxb&=~(0x04<<5) if($dst>=8);
852 $rxb&=~(0x01<<5) if($src1>=8);
853 $rxb&=~(0x02<<5) if($src2>=8);
858 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
860 rxb(\@opcode,$3,$2,-1,0x08);
861 push @opcode,0x78,0xc2;
862 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
864 push @opcode,$c=~/^0/?oct($c):$c;
872 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
874 rxb(\@opcode,$3,$2,-1,0x08);
875 push @opcode,0x78,0xc3;
876 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
878 push @opcode,$c=~/^0/?oct($c):$c;
885 my $endbranch = sub {
886 (0xf3,0x0f,0x1e,0xfa);
901 while(defined(my $line=<>)) {
903 $line =~ s|\R$||; # Better chomp
905 $line =~ s|[#!].*$||; # get rid of asm-style comments...
906 $line =~ s|/\*.*\*/||; # ... and C-style comments...
907 $line =~ s|^\s+||; # ... and skip white spaces in beginning
908 $line =~ s|\s+$||; # ... and at the end
910 if (my $label=label->re(\$line)) { print $label->out(); }
912 if (my $directive=directive->re(\$line)) {
913 printf "%s",$directive->out();
914 } elsif (my $opcode=opcode->re(\$line)) {
915 my $asm = eval("\$".$opcode->mnemonic());
917 if ((ref($asm) eq 'CODE') && scalar(my @bytes=&$asm($line))) {
918 print $gas?".byte\t":"DB\t",join(',',@bytes),"\n";
923 ARGUMENT: while (1) {
926 ($arg=register->re(\$line, $opcode))||
927 ($arg=const->re(\$line)) ||
928 ($arg=ea->re(\$line, $opcode)) ||
929 ($arg=expr->re(\$line, $opcode)) ||
934 last ARGUMENT if ($line !~ /^,/);
941 my $sz=$opcode->size();
944 $insn = $opcode->out($#args>=1?$args[$#args]->size():$sz);
945 @args = map($_->out($sz),@args);
946 printf "\t%s\t%s",$insn,join(",",@args);
948 $insn = $opcode->out();
951 # $insn.=$sz compensates for movq, pinsrw, ...
952 if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; }
953 if ($arg =~ /^ymm[0-9]+$/) { $insn.=$sz; $sz="y" if(!$sz); last; }
954 if ($arg =~ /^zmm[0-9]+$/) { $insn.=$sz; $sz="z" if(!$sz); last; }
955 if ($arg =~ /^mm[0-9]+$/) { $insn.=$sz; $sz="q" if(!$sz); last; }
957 @args = reverse(@args);
958 undef $sz if ($nasm && $opcode->mnemonic() eq "lea");
959 printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args));
962 printf "\t%s",$opcode->out();
969 print "\n$current_segment\tENDS\n" if ($current_segment && $masm);
970 print "END\n" if ($masm);
974 \f#################################################
975 # Cross-reference x86_64 ABI "card"
995 # (*) volatile register
996 # (-) preserved by callee
997 # (#) Nth argument, volatile
999 # In Unix terms top of stack is argument transfer area for arguments
1000 # which could not be accommodated in registers. Or in other words 7th
1001 # [integer] argument resides at 8(%rsp) upon function entry point.
1002 # 128 bytes above %rsp constitute a "red zone" which is not touched
1003 # by signal handlers and can be used as temporal storage without
1004 # allocating a frame.
1006 # In Win64 terms N*8 bytes on top of stack is argument transfer area,
1007 # which belongs to/can be overwritten by callee. N is the number of
1008 # arguments passed to callee, *but* not less than 4! This means that
1009 # upon function entry point 5th argument resides at 40(%rsp), as well
1010 # as that 32 bytes from 8(%rsp) can always be used as temporal
1011 # storage [without allocating a frame]. One can actually argue that
1012 # one can assume a "red zone" above stack pointer under Win64 as well.
1013 # Point is that at apparently no occasion Windows kernel would alter
1014 # the area above user stack pointer in true asynchronous manner...
1016 # All the above means that if assembler programmer adheres to Unix
1017 # register and stack layout, but disregards the "red zone" existense,
1018 # it's possible to use following prologue and epilogue to "gear" from
1019 # Unix to Win64 ABI in leaf functions with not more than 6 arguments.
1021 # omnipotent_function:
1024 # movq %rsi,16(%rsp)
1025 # movq %rcx,%rdi ; if 1st argument is actually present
1026 # movq %rdx,%rsi ; if 2nd argument is actually ...
1027 # movq %r8,%rdx ; if 3rd argument is ...
1028 # movq %r9,%rcx ; if 4th argument ...
1029 # movq 40(%rsp),%r8 ; if 5th ...
1030 # movq 48(%rsp),%r9 ; if 6th ...
1035 # movq 16(%rsp),%rsi
1039 \f#################################################
1040 # Win64 SEH, Structured Exception Handling.
1042 # Unlike on Unix systems(*) lack of Win64 stack unwinding information
1043 # has undesired side-effect at run-time: if an exception is raised in
1044 # assembler subroutine such as those in question (basically we're
1045 # referring to segmentation violations caused by malformed input
1046 # parameters), the application is briskly terminated without invoking
1047 # any exception handlers, most notably without generating memory dump
1048 # or any user notification whatsoever. This poses a problem. It's
1049 # possible to address it by registering custom language-specific
1050 # handler that would restore processor context to the state at
1051 # subroutine entry point and return "exception is not handled, keep
1052 # unwinding" code. Writing such handler can be a challenge... But it's
1053 # doable, though requires certain coding convention. Consider following
1056 # .type function,@function
1058 # movq %rsp,%rax # copy rsp to volatile register
1059 # pushq %r15 # save non-volatile registers
1063 # subq %rdi,%r11 # prepare [variable] stack frame
1065 # movq %rax,0(%r11) # check for exceptions
1066 # movq %r11,%rsp # allocate [variable] stack frame
1067 # movq %rax,0(%rsp) # save original rsp value
1070 # movq 0(%rsp),%rcx # pull original rsp value
1071 # movq -24(%rcx),%rbp # restore non-volatile registers
1072 # movq -16(%rcx),%rbx
1073 # movq -8(%rcx),%r15
1074 # movq %rcx,%rsp # restore original rsp
1077 # .size function,.-function
1079 # The key is that up to magic_point copy of original rsp value remains
1080 # in chosen volatile register and no non-volatile register, except for
1081 # rsp, is modified. While past magic_point rsp remains constant till
1082 # the very end of the function. In this case custom language-specific
1083 # exception handler would look like this:
1085 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1086 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1087 # { ULONG64 *rsp = (ULONG64 *)context->Rax;
1088 # ULONG64 rip = context->Rip;
1090 # if (rip >= magic_point)
1091 # { rsp = (ULONG64 *)context->Rsp;
1092 # if (rip < magic_epilogue)
1093 # { rsp = (ULONG64 *)rsp[0];
1094 # context->Rbp = rsp[-3];
1095 # context->Rbx = rsp[-2];
1096 # context->R15 = rsp[-1];
1099 # context->Rsp = (ULONG64)rsp;
1100 # context->Rdi = rsp[1];
1101 # context->Rsi = rsp[2];
1103 # memcpy (disp->ContextRecord,context,sizeof(CONTEXT));
1104 # RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase,
1105 # dips->ControlPc,disp->FunctionEntry,disp->ContextRecord,
1106 # &disp->HandlerData,&disp->EstablisherFrame,NULL);
1107 # return ExceptionContinueSearch;
1110 # It's appropriate to implement this handler in assembler, directly in
1111 # function's module. In order to do that one has to know members'
1112 # offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant
1113 # values. Here they are:
1133 # sizeof(CONTEXT) 1232
1134 # DISPATCHER_CONTEXT.ControlPc 0
1135 # DISPATCHER_CONTEXT.ImageBase 8
1136 # DISPATCHER_CONTEXT.FunctionEntry 16
1137 # DISPATCHER_CONTEXT.EstablisherFrame 24
1138 # DISPATCHER_CONTEXT.TargetIp 32
1139 # DISPATCHER_CONTEXT.ContextRecord 40
1140 # DISPATCHER_CONTEXT.LanguageHandler 48
1141 # DISPATCHER_CONTEXT.HandlerData 56
1142 # UNW_FLAG_NHANDLER 0
1143 # ExceptionContinueSearch 1
1145 # In order to tie the handler to the function one has to compose
1146 # couple of structures: one for .xdata segment and one for .pdata.
1148 # UNWIND_INFO structure for .xdata segment would be
1150 # function_unwind_info:
1154 # This structure designates exception handler for a function with
1155 # zero-length prologue, no stack frame or frame register.
1157 # To facilitate composing of .pdata structures, auto-generated "gear"
1158 # prologue copies rsp value to rax and denotes next instruction with
1159 # .LSEH_begin_{function_name} label. This essentially defines the SEH
1160 # styling rule mentioned in the beginning. Position of this label is
1161 # chosen in such manner that possible exceptions raised in the "gear"
1162 # prologue would be accounted to caller and unwound from latter's frame.
1163 # End of function is marked with respective .LSEH_end_{function_name}
1164 # label. To summarize, .pdata segment would contain
1166 # .rva .LSEH_begin_function
1167 # .rva .LSEH_end_function
1168 # .rva function_unwind_info
1170 # Reference to function_unwind_info from .xdata segment is the anchor.
1171 # In case you wonder why references are 32-bit .rvas and not 64-bit
1172 # .quads. References put into these two segments are required to be
1173 # *relative* to the base address of the current binary module, a.k.a.
1174 # image base. No Win64 module, be it .exe or .dll, can be larger than
1175 # 2GB and thus such relative references can be and are accommodated in
1178 # Having reviewed the example function code, one can argue that "movq
1179 # %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix
1180 # rax would contain an undefined value. If this "offends" you, use
1181 # another register and refrain from modifying rax till magic_point is
1182 # reached, i.e. as if it was a non-volatile register. If more registers
1183 # are required prior [variable] frame setup is completed, note that
1184 # nobody says that you can have only one "magic point." You can
1185 # "liberate" non-volatile registers by denoting last stack off-load
1186 # instruction and reflecting it in finer grade unwind logic in handler.
1187 # After all, isn't it why it's called *language-specific* handler...
1189 # SE handlers are also involved in unwinding stack when executable is
1190 # profiled or debugged. Profiling implies additional limitations that
1191 # are too subtle to discuss here. For now it's sufficient to say that
1192 # in order to simplify handlers one should either a) offload original
1193 # %rsp to stack (like discussed above); or b) if you have a register to
1194 # spare for frame pointer, choose volatile one.
1196 # (*) Note that we're talking about run-time, not debug-time. Lack of
1197 # unwind information makes debugging hard on both Windows and
1198 # Unix. "Unlike" referes to the fact that on Unix signal handler
1199 # will always be invoked, core dumped and appropriate exit code
1200 # returned to parent (for user notification).