3 # Ascetic x86_64 AT&T to MASM assembler translator by <appro>.
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...
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.
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...
20 # Currently recognized limitations:
22 # - can't use multiple ops per line;
23 # - indirect calls and jumps are not supported;
25 # Dual-ABI styling rules.
27 # 1. Adhere to Unix register and stack layout [see the end for
29 # 2. Forget about "red zone," stick to more traditional blended
30 # stack frame allocation. If volatile storage is actually required
31 # that is. If not, just leave the stack as is.
32 # 3. Functions tagged with ".type name,@function" get crafted with
33 # unified Win64 prologue and epilogue automatically. If you want
34 # to take care of ABI differences yourself, tag functions as
35 # ".type name,@abi-omnipotent" instead.
36 # 4. To optimize the Win64 prologue you can specify number of input
37 # arguments as ".type name,@function,N." Keep in mind that if N is
38 # larger than 6, then you *have to* write "abi-omnipotent" code,
39 # because >6 cases can't be addressed with unified prologue.
40 # 5. Name local labels as .L*, do *not* use dynamic labels such as 1:
41 # (sorry about latter).
42 # 6. Don't use [or hand-code with .byte] "rep ret." "ret" mnemonic is
43 # required to identify the spots, where to inject Win64 epilogue!
44 # But on the pros, it's then prefixed with rep automatically:-)
47 open STDOUT,">$output" || die "can't open $output: $!";
49 my $masm=1 if ($output =~ /\.asm/);
54 { package opcode; # pick up opcodes
56 my $self = shift; # single instance in enough...
60 if ($line =~ /^([a-z]+)/i) {
63 $line = substr($line,@+[0]); $line =~ s/^\s+//;
66 if ($self->{op} =~ /(movz)b.*/) { # movz is pain...
69 } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])/) {
79 $self->{sz} = $sz if (defined($sz) && !defined($self->{sz}));
85 if ($self->{op} eq "movz") { # movz in pain...
86 sprintf "%s%s%s",$self->{op},$self->{sz},shift;
87 } elsif ($self->{op} eq "ret") {
90 "$self->{op}$self->{sz}";
93 $self->{op} =~ s/movz/movzx/;
94 if ($self->{op} eq "ret") {
96 if ($current_function->{abi} eq "svr4") {
97 $self->{op} = "mov rdi,QWORD PTR 8[rsp]\t;WIN64 epilogue\n\t".
98 "mov rsi,QWORD PTR 16[rsp]\n\t";
100 $self->{op} .= "DB\t0F3h,0C3h\t\t;repret";
106 { package const; # pick up constants, which start with $
108 my $self = shift; # single instance in enough...
112 if ($line =~ /^\$([^,]+)/) {
115 $line = substr($line,@+[0]); $line =~ s/^\s+//;
117 $self->{value} = oct($self->{value}) if ($self->{value} =~ /^0/);
123 sprintf $masm?"0%xh":"\$0x%x",$self->{value};
126 { package ea; # pick up effective addresses: expr(%reg,%reg,scale)
128 my $self = shift; # single instance in enough...
132 if ($line =~ /^([^\(,]*)\(([%\w,]+)\)/) {
134 ($self->{base},$self->{index},$self->{scale})=split(/,/,$2);
135 $self->{scale} = 1 if (!defined($self->{scale}));
137 $line = substr($line,@+[0]); $line =~ s/^\s+//;
139 $self->{base} =~ s/^%//;
140 $self->{index} =~ s/^%// if (defined($self->{index}));
150 # elder GNU assembler insists on 64-bit EAs:-(
151 # on pros side, this results in more compact code:-)
152 $self->{index} =~ s/^[er](.?[0-9xp])[d]?$/r\1/;
153 $self->{base} =~ s/^[er](.?[0-9xp])[d]?$/r\1/;
155 if (defined($self->{index})) {
156 sprintf "%s(%%%s,%%%s,%d)",
157 $self->{label},$self->{base},
158 $self->{index},$self->{scale};
161 sprintf "%s(%%%s)", $self->{label},$self->{base};
164 %szmap = ( b=>"BYTE", w=>"WORD", l=>"DWORD", q=>"QWORD" );
166 $self->{label} =~ s/\./\$/g;
167 $self->{label} =~ s/0x([0-9a-f]+)/0$1h/ig;
168 $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/);
170 if (defined($self->{index})) {
171 sprintf "%s PTR %s[%s*%d+%s]",$szmap{$sz},
173 $self->{index},$self->{scale},
177 sprintf "%s PTR %s[%s]",$szmap{$sz},
178 $self->{label},$self->{base};
183 { package register; # pick up registers, which start with %.
185 my $class = shift; # muliple instances...
190 if ($line =~ /^%(\w+)/) {
194 $line = substr($line,@+[0]); $line =~ s/^\s+//;
202 if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; }
203 elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; }
204 elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; }
205 elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; }
206 elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; }
207 elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; }
208 elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; }
209 elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; }
215 sprintf $masm?"%s":"%%%s",$self->{value};
218 { package label; # pick up labels, which end with :
220 my $self = shift; # single instance is enough...
224 if ($line =~ /(^[\.\w]+\:)/) {
227 $line = substr($line,@+[0]); $line =~ s/^\s+//;
229 $self->{value} =~ s/\.L/\$L/ if ($masm);
238 } elsif ($self->{value} ne "$current_function->{name}:") {
240 } elsif ($current_function->{abi} eq "svr4") {
241 my $func = "$current_function->{name} PROC\n".
242 " mov QWORD PTR 8[rsp],rdi\t;WIN64 prologue\n".
243 " mov QWORD PTR 16[rsp],rsi\n";
244 my $narg = $current_function->{narg};
245 $narg=6 if (!defined($narg));
246 $func .= " mov rdi,rcx\n" if ($narg>0);
247 $func .= " mov rsi,rdx\n" if ($narg>1);
248 $func .= " mov rdx,r8\n" if ($narg>2);
249 $func .= " mov rcx,r9\n" if ($narg>3);
250 $func .= " mov r8,QWORD PTR 40[rsp]\n" if ($narg>4);
251 $func .= " mov r9,QWORD PTR 48[rsp]\n" if ($narg>5);
254 "$current_function->{name} PROC";
258 { package expr; # pick up expressioins
260 my $self = shift; # single instance is enough...
264 if ($line =~ /(^[^,]+)/) {
267 $line = substr($line,@+[0]); $line =~ s/^\s+//;
269 $self->{value} =~ s/\.L/\$L/g if ($masm);
278 { package directive; # pick up directives, which start with .
280 my $self = shift; # single instance is enough...
285 if ($line =~ /^\s*(\.\w+)/) {
288 $line =~ s/\@abi\-omnipotent/\@function/;
289 $line =~ s/\@function.*/\@function/;
290 $self->{value} = $line;
297 undef $self->{value};
298 $line = substr($line,@+[0]); $line =~ s/^\s+//;
302 $v="$current_segment\tENDS\n" if ($current_segment);
303 $current_segment = "_$1";
304 $current_segment =~ tr/[a-z]/[A-Z]/;
305 $v.="$current_segment\tSEGMENT PARA";
309 /\.globl/ && do { $self->{value} = "PUBLIC\t".$line; last; };
310 /\.type/ && do { ($sym,$type,$narg) = split(',',$line);
311 if ($type eq "\@function")
312 { undef $current_function;
313 $current_function->{name} = $sym;
314 $current_function->{abi} = "svr4";
315 $current_function->{narg} = $narg;
317 elsif ($type eq "\@abi-omnipotent")
318 { undef $current_function;
319 $current_function->{name} = $sym;
323 /\.size/ && do { if (defined($current_function))
324 { $self->{value}="$current_function->{name}\tENDP";
325 undef $current_function;
329 /\.align/ && do { $self->{value} = "ALIGN\t".$line; last; };
330 /\.(byte|value|long|quad)/
331 && do { my @arr = split(',',$line);
332 my $sz = substr($1,0,1);
333 my $last = pop(@arr);
335 $sz =~ tr/bvlq/BWDQ/;
336 $self->{value} = "\tD$sz\t";
337 for (@arr) { $self->{value} .= sprintf"0%Xh,",oct; }
338 $self->{value} .= sprintf"0%Xh",oct($last);
357 $line =~ s|[#!].*$||; # get rid of asm-style comments...
358 $line =~ s|/\*.*\*/||; # ... and C-style comments...
359 $line =~ s|^\s+||; # ... and skip white spaces in beginning
367 if ($label=label->re(\$line)) { print $label->out(); }
369 if (directive->re(\$line)) {
370 printf "%s",directive->out();
371 } elsif ($opcode=opcode->re(\$line)) { ARGUMENT: {
373 if ($src=register->re(\$line)) { opcode->size($src->size()); }
374 elsif ($src=const->re(\$line)) { }
375 elsif ($src=ea->re(\$line)) { }
376 elsif ($src=expr->re(\$line)) { }
378 last ARGUMENT if ($line !~ /^,/);
380 $line = substr($line,1); $line =~ s/^\s+//;
382 if ($dst=register->re(\$line)) { opcode->size($dst->size()); }
383 elsif ($dst=const->re(\$line)) { }
384 elsif ($dst=ea->re(\$line)) { }
392 printf "\t%s\t%s,%s", $opcode->out($dst->size()),
393 $src->out($sz),$dst->out($sz);
396 printf "\t%s\t%s,%s", $opcode->out(),
397 $dst->out($sz),$src->out($sz);
400 elsif (defined($src)) {
401 printf "\t%s\t%s",$opcode->out(),$src->out($sz);
403 printf "\t%s",$opcode->out();
410 print "\n$current_segment\tENDS\nEND\n" if ($masm);
414 #################################################
415 # Cross-reference x86_64 ABI "card"
435 # (*) volatile register
436 # (-) preserved by callee
437 # (#) Nth argument, volatile
439 # In Unix terms top of stack is argument transfer area for arguments
440 # which could not be accomodated in registers. Or in other words 7th
441 # [integer] argument resides at 8(%rsp) upon function entry point.
442 # 128 bytes above %rsp constitute a "red zone" which is not touched
443 # by signal handlers and can be used as temporal storage without
444 # allocating a frame.
446 # In Win64 terms N*8 bytes on top of stack is argument transfer area,
447 # which belongs to/can be overwritten by callee. N is the number of
448 # arguments passed to callee, *but* not less than 4! This means that
449 # upon function entry point 5th argument resides at 40(%rsp), as well
450 # as that 32 bytes from 8(%rsp) can always be used as temporal
451 # storage [without allocating a frame].
453 # All the above means that if assembler programmer adheres to Unix
454 # register and stack layout, but disregards the "red zone" existense,
455 # it's possible to use following prologue and epilogue to "gear" from
456 # Unix to Win64 ABI in leaf functions with not more than 6 arguments.
458 # omnipotent_function:
462 # movq %rcx,%rdi ; if 1st argument is actually present
463 # movq %rdx,%rsi ; if 2nd argument is actually ...
464 # movq %r8,%rdx ; if 3rd argument is ...
465 # movq %r9,%rcx ; if 4th argument ...
466 # movq 40(%rsp),%r8 ; if 5th ...
467 # movq 48(%rsp),%r9 ; if 6th ...