]>
Commit | Line | Data |
---|---|---|
1cfd258e AP |
1 | #!/usr/bin/env perl |
2 | ||
80aa9cc9 | 3 | # Ascetic x86_64 AT&T to MASM/NASM assembler translator by <appro>. |
1cfd258e AP |
4 | # |
5 | # Why AT&T to MASM and not vice versa? Several reasons. Because AT&T | |
6 | # format is way easier to parse. Because it's simpler to "gear" from | |
7 | # Unix ABI to Windows one [see cross-reference "card" at the end of | |
8 | # file]. Because Linux targets were available first... | |
9 | # | |
10 | # In addition the script also "distills" code suitable for GNU | |
11 | # assembler, so that it can be compiled with more rigid assemblers, | |
12 | # such as Solaris /usr/ccs/bin/as. | |
13 | # | |
14 | # This translator is not designed to convert *arbitrary* assembler | |
15 | # code from AT&T format to MASM one. It's designed to convert just | |
16 | # enough to provide for dual-ABI OpenSSL modules development... | |
17 | # There *are* limitations and you might have to modify your assembler | |
18 | # code or this script to achieve the desired result... | |
19 | # | |
20 | # Currently recognized limitations: | |
21 | # | |
22 | # - can't use multiple ops per line; | |
1cfd258e AP |
23 | # |
24 | # Dual-ABI styling rules. | |
25 | # | |
80aa9cc9 AP |
26 | # 1. Adhere to Unix register and stack layout [see cross-reference |
27 | # ABI "card" at the end for explanation]. | |
1cfd258e AP |
28 | # 2. Forget about "red zone," stick to more traditional blended |
29 | # stack frame allocation. If volatile storage is actually required | |
30 | # that is. If not, just leave the stack as is. | |
31 | # 3. Functions tagged with ".type name,@function" get crafted with | |
4b450519 | 32 | # unified Win64 prologue and epilogue automatically. If you want |
1cfd258e | 33 | # to take care of ABI differences yourself, tag functions as |
4b450519 AP |
34 | # ".type name,@abi-omnipotent" instead. |
35 | # 4. To optimize the Win64 prologue you can specify number of input | |
36 | # arguments as ".type name,@function,N." Keep in mind that if N is | |
37 | # larger than 6, then you *have to* write "abi-omnipotent" code, | |
38 | # because >6 cases can't be addressed with unified prologue. | |
39 | # 5. Name local labels as .L*, do *not* use dynamic labels such as 1: | |
40 | # (sorry about latter). | |
41 | # 6. Don't use [or hand-code with .byte] "rep ret." "ret" mnemonic is | |
42 | # required to identify the spots, where to inject Win64 epilogue! | |
43 | # But on the pros, it's then prefixed with rep automatically:-) | |
85253772 AP |
44 | # 7. Stick to explicit ip-relative addressing. If you have to use |
45 | # GOTPCREL addressing, stick to mov symbol@GOTPCREL(%rip),%r??. | |
46 | # Both are recognized and translated to proper Win64 addressing | |
47 | # modes. To support legacy code a synthetic directive, .picmeup, | |
48 | # is implemented. It puts address of the *next* instruction into | |
49 | # target register, e.g.: | |
5d0d60e2 | 50 | # |
5d0d60e2 AP |
51 | # .picmeup %rax |
52 | # lea .Label-.(%rax),%rax | |
80aa9cc9 AP |
53 | # |
54 | # 8. In order to provide for structured exception handling unified | |
55 | # Win64 prologue copies %rsp value to %rax. For further details | |
56 | # see SEH paragraph at the end. | |
85253772 | 57 | # 9. .init segment is allowed to contain calls to functions only. |
80aa9cc9 | 58 | \f |
85253772 AP |
59 | my $flavour = shift; |
60 | my $output = shift; | |
61 | if ($flavour =~ /\./) { $output = $flavour; undef $flavour; } | |
932cc129 AP |
62 | |
63 | { my ($stddev,$stdino,@junk)=stat(STDOUT); | |
64 | my ($outdev,$outino,@junk)=stat($output); | |
65 | ||
66 | open STDOUT,">$output" || die "can't open $output: $!" | |
67 | if ($stddev!=$outdev || $stdino!=$outino); | |
68 | } | |
1cfd258e | 69 | |
85253772 AP |
70 | my $gas=1; $gas=0 if ($output =~ /\.asm$/); |
71 | my $elf=1; $elf=0 if (!$gas); | |
72 | my $win64=0; | |
73 | my $prefix=""; | |
a23e3dbe | 74 | |
8ab9025e | 75 | my $masmref=8 + 50727*2**-32; # 8.00.50727 shipped with VS2005 |
9b634c9b | 76 | my $masm=0; |
a23e3dbe AP |
77 | my $PTR=" PTR"; |
78 | ||
9b634c9b | 79 | my $nasmref=2.03; |
a23e3dbe AP |
80 | my $nasm=0; |
81 | ||
85253772 AP |
82 | if ($flavour eq "mingw64") { $gas=1; $elf=0; $win64=1; $prefix="_"; } |
83 | elsif ($flavour eq "macosx") { $gas=1; $elf=0; $prefix="_"; } | |
84 | elsif ($flavour eq "masm") { $gas=0; $elf=0; $masm=$masmref; $win64=1; } | |
85 | elsif ($flavour eq "nasm") { $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $PTR=""; } | |
86 | elsif (!$gas) | |
9b634c9b AP |
87 | { if ($ENV{ASM} =~ m/nasm/ && `nasm -v` =~ m/version ([0-9]+)\.([0-9]+)/i) |
88 | { $nasm = $1 + $2*0.01; $PTR=""; } | |
a23e3dbe AP |
89 | elsif (`ml64 2>&1` =~ m/Version ([0-9]+)\.([0-9]+)(\.([0-9]+))?/) |
90 | { $masm = $1 + $2*2**-16 + $4*2**-32; } | |
9b634c9b | 91 | die "no assembler found on %PATH" if (!($nasm || $masm)); |
85253772 AP |
92 | $win64=1; |
93 | $elf=0; | |
a23e3dbe | 94 | } |
1cfd258e AP |
95 | |
96 | my $current_segment; | |
97 | my $current_function; | |
9b634c9b | 98 | my %globals; |
1cfd258e AP |
99 | |
100 | { package opcode; # pick up opcodes | |
101 | sub re { | |
102 | my $self = shift; # single instance in enough... | |
103 | local *line = shift; | |
104 | undef $ret; | |
105 | ||
932cc129 | 106 | if ($line =~ /^([a-z][a-z0-9]*)/i) { |
1cfd258e AP |
107 | $self->{op} = $1; |
108 | $ret = $self; | |
109 | $line = substr($line,@+[0]); $line =~ s/^\s+//; | |
110 | ||
111 | undef $self->{sz}; | |
a078befc | 112 | if ($self->{op} =~ /^(movz)b.*/) { # movz is pain... |
1cfd258e AP |
113 | $self->{op} = $1; |
114 | $self->{sz} = "b"; | |
9b634c9b | 115 | } elsif ($self->{op} =~ /call|jmp/) { |
55eab3b7 | 116 | $self->{sz} = "" |
a078befc | 117 | } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) { |
1cfd258e AP |
118 | $self->{op} = $1; |
119 | $self->{sz} = $2; | |
120 | } | |
121 | } | |
122 | $ret; | |
123 | } | |
124 | sub size { | |
125 | my $self = shift; | |
126 | my $sz = shift; | |
127 | $self->{sz} = $sz if (defined($sz) && !defined($self->{sz})); | |
128 | $self->{sz}; | |
129 | } | |
130 | sub out { | |
131 | my $self = shift; | |
85253772 | 132 | if ($gas) { |
932cc129 | 133 | if ($self->{op} eq "movz") { # movz is pain... |
1cfd258e | 134 | sprintf "%s%s%s",$self->{op},$self->{sz},shift; |
932cc129 AP |
135 | } elsif ($self->{op} =~ /^set/) { |
136 | "$self->{op}"; | |
1cfd258e | 137 | } elsif ($self->{op} eq "ret") { |
85253772 AP |
138 | my $epilogue = ""; |
139 | if ($win64 && $current_function->{abi} eq "svr4") { | |
140 | $epilogue = "movq 8(%rsp),%rdi\n\t" . | |
141 | "movq 16(%rsp),%rsi\n\t"; | |
142 | } | |
143 | $epilogue . ".byte 0xf3,0xc3"; | |
144 | } elsif ($self->{op} eq "call" && !$elf && $current_segment eq ".init") { | |
145 | ".p2align\t3\n\t.quad"; | |
1cfd258e AP |
146 | } else { |
147 | "$self->{op}$self->{sz}"; | |
148 | } | |
149 | } else { | |
a078befc | 150 | $self->{op} =~ s/^movz/movzx/; |
1cfd258e AP |
151 | if ($self->{op} eq "ret") { |
152 | $self->{op} = ""; | |
85253772 | 153 | if ($win64 && $current_function->{abi} eq "svr4") { |
a23e3dbe AP |
154 | $self->{op} = "mov rdi,QWORD${PTR}[8+rsp]\t;WIN64 epilogue\n\t". |
155 | "mov rsi,QWORD${PTR}[16+rsp]\n\t"; | |
1cfd258e AP |
156 | } |
157 | $self->{op} .= "DB\t0F3h,0C3h\t\t;repret"; | |
85253772 | 158 | } elsif ($self->{op} =~ /^(pop|push)f/) { |
80aa9cc9 | 159 | $self->{op} .= $self->{sz}; |
85253772 AP |
160 | } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") { |
161 | $self->{op} = "ALIGN\t8\n\tDQ"; | |
162 | } | |
1cfd258e AP |
163 | $self->{op}; |
164 | } | |
165 | } | |
85253772 AP |
166 | sub mnemonic { |
167 | my $self=shift; | |
168 | my $op=shift; | |
169 | $self->{op}=$op if (defined($op)); | |
170 | $self->{op}; | |
171 | } | |
1cfd258e AP |
172 | } |
173 | { package const; # pick up constants, which start with $ | |
174 | sub re { | |
175 | my $self = shift; # single instance in enough... | |
176 | local *line = shift; | |
177 | undef $ret; | |
178 | ||
179 | if ($line =~ /^\$([^,]+)/) { | |
180 | $self->{value} = $1; | |
181 | $ret = $self; | |
182 | $line = substr($line,@+[0]); $line =~ s/^\s+//; | |
183 | } | |
184 | $ret; | |
185 | } | |
186 | sub out { | |
187 | my $self = shift; | |
5d0d60e2 | 188 | |
85253772 | 189 | if ($gas) { |
e84b663a AP |
190 | # Solaris /usr/ccs/bin/as can't handle multiplications |
191 | # in $self->{value} | |
192 | $self->{value} =~ s/(?<![0-9a-f])(0[x0-9a-f]+)/oct($1)/egi; | |
193 | $self->{value} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg; | |
5d0d60e2 AP |
194 | sprintf "\$%s",$self->{value}; |
195 | } else { | |
196 | $self->{value} =~ s/0x([0-9a-f]+)/0$1h/ig; | |
197 | sprintf "%s",$self->{value}; | |
198 | } | |
1cfd258e AP |
199 | } |
200 | } | |
201 | { package ea; # pick up effective addresses: expr(%reg,%reg,scale) | |
202 | sub re { | |
203 | my $self = shift; # single instance in enough... | |
204 | local *line = shift; | |
205 | undef $ret; | |
206 | ||
9b634c9b AP |
207 | # optional * ---vvv--- appears in indirect jmp/call |
208 | if ($line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)/) { | |
209 | $self->{asterisk} = $1; | |
210 | $self->{label} = $2; | |
211 | ($self->{base},$self->{index},$self->{scale})=split(/,/,$3); | |
1cfd258e AP |
212 | $self->{scale} = 1 if (!defined($self->{scale})); |
213 | $ret = $self; | |
214 | $line = substr($line,@+[0]); $line =~ s/^\s+//; | |
215 | ||
85253772 AP |
216 | if ($win64 && $self->{label} =~ s/\@GOTPCREL//) { |
217 | die if (opcode->mnemonic() ne "mov"); | |
218 | opcode->mnemonic("lea"); | |
219 | } | |
1cfd258e AP |
220 | $self->{base} =~ s/^%//; |
221 | $self->{index} =~ s/^%// if (defined($self->{index})); | |
222 | } | |
223 | $ret; | |
224 | } | |
225 | sub size {} | |
226 | sub out { | |
227 | my $self = shift; | |
228 | my $sz = shift; | |
229 | ||
85253772 AP |
230 | $self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; |
231 | ||
afbe674e AP |
232 | # Silently convert all EAs to 64-bit. This is required for |
233 | # elder GNU assembler and results in more compact code, | |
234 | # *but* most importantly AES module depends on this feature! | |
235 | $self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; | |
236 | $self->{base} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; | |
237 | ||
85253772 | 238 | if ($gas) { |
e442c362 AP |
239 | # Solaris /usr/ccs/bin/as can't handle multiplications |
240 | # in $self->{label} | |
e84b663a | 241 | $self->{label} =~ s/(?<![0-9a-f])(0[x0-9a-f]+)/oct($1)/egi; |
e442c362 | 242 | $self->{label} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg; |
85253772 | 243 | $self->{label} =~ s/\.L/L\$/g if (!$elf); |
4b450519 | 244 | |
1cfd258e | 245 | if (defined($self->{index})) { |
9b634c9b | 246 | sprintf "%s%s(%%%s,%%%s,%d)",$self->{asterisk}, |
70cf3095 | 247 | $self->{label},$self->{base}, |
1cfd258e | 248 | $self->{index},$self->{scale}; |
5d0d60e2 | 249 | } else { |
9b634c9b | 250 | sprintf "%s%s(%%%s)", $self->{asterisk},$self->{label},$self->{base}; |
1cfd258e AP |
251 | } |
252 | } else { | |
a23e3dbe | 253 | %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR", l=>"DWORD$PTR", q=>"QWORD$PTR" ); |
1cfd258e | 254 | |
80aa9cc9 | 255 | $self->{label} =~ s/\.L/\$L\$/g; |
70cf3095 AP |
256 | $self->{label} =~ s/\./\$/g; |
257 | $self->{label} =~ s/0x([0-9a-f]+)/0$1h/ig; | |
258 | $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/); | |
9b634c9b | 259 | $sz="q" if ($self->{asterisk}); |
70cf3095 | 260 | |
1cfd258e | 261 | if (defined($self->{index})) { |
a23e3dbe AP |
262 | sprintf "%s[%s%s*%d+%s]",$szmap{$sz}, |
263 | $self->{label}?"$self->{label}+":"", | |
1cfd258e AP |
264 | $self->{index},$self->{scale}, |
265 | $self->{base}; | |
932cc129 | 266 | } elsif ($self->{base} eq "rip") { |
a23e3dbe | 267 | sprintf "%s[%s]",$szmap{$sz},$self->{label}; |
5d0d60e2 | 268 | } else { |
a23e3dbe AP |
269 | sprintf "%s[%s%s]",$szmap{$sz}, |
270 | $self->{label}?"$self->{label}+":"", | |
271 | $self->{base}; | |
1cfd258e AP |
272 | } |
273 | } | |
274 | } | |
275 | } | |
276 | { package register; # pick up registers, which start with %. | |
277 | sub re { | |
278 | my $class = shift; # muliple instances... | |
279 | my $self = {}; | |
280 | local *line = shift; | |
281 | undef $ret; | |
282 | ||
9b634c9b AP |
283 | # optional * ---vvv--- appears in indirect jmp/call |
284 | if ($line =~ /^(\*?)%(\w+)/) { | |
1cfd258e | 285 | bless $self,$class; |
9b634c9b AP |
286 | $self->{asterisk} = $1; |
287 | $self->{value} = $2; | |
1cfd258e AP |
288 | $ret = $self; |
289 | $line = substr($line,@+[0]); $line =~ s/^\s+//; | |
290 | } | |
291 | $ret; | |
292 | } | |
293 | sub size { | |
294 | my $self = shift; | |
295 | undef $ret; | |
296 | ||
297 | if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; } | |
298 | elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; } | |
299 | elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; } | |
300 | elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; } | |
301 | elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; } | |
302 | elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; } | |
303 | elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; } | |
304 | elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; } | |
305 | ||
306 | $ret; | |
307 | } | |
308 | sub out { | |
309 | my $self = shift; | |
85253772 | 310 | if ($gas) { sprintf "%s%%%s",$self->{asterisk},$self->{value}; } |
9b634c9b | 311 | else { $self->{value}; } |
1cfd258e AP |
312 | } |
313 | } | |
314 | { package label; # pick up labels, which end with : | |
315 | sub re { | |
316 | my $self = shift; # single instance is enough... | |
317 | local *line = shift; | |
318 | undef $ret; | |
319 | ||
85253772 | 320 | if ($line =~ /(^[\.\w]+)\:/) { |
1cfd258e AP |
321 | $self->{value} = $1; |
322 | $ret = $self; | |
323 | $line = substr($line,@+[0]); $line =~ s/^\s+//; | |
324 | ||
85253772 AP |
325 | $self->{value} =~ s/\.L/\$L\$/ if (!$gas); |
326 | $self->{value} =~ s/\.L/L\$/ if (!$elf); | |
1cfd258e AP |
327 | } |
328 | $ret; | |
329 | } | |
330 | sub out { | |
331 | my $self = shift; | |
332 | ||
85253772 AP |
333 | if ($gas) { |
334 | my $func = ($globals{$self->{value}} or $self->{value}) . ":"; | |
335 | if ($win64 && | |
336 | $current_function->{name} eq $self->{value} && | |
337 | $current_function->{abi} eq "svr4") { | |
338 | $func .= "\n"; | |
339 | $func .= " movq %rdi,8(%rsp)\n"; | |
340 | $func .= " movq %rsi,16(%rsp)\n"; | |
341 | $func .= " movq %rsp,%rax\n"; | |
342 | $func .= ".LSEH_begin_$current_function->{name}:\n"; | |
343 | my $narg = $current_function->{narg}; | |
344 | $narg=6 if (!defined($narg)); | |
345 | $func .= " movq %rcx,%rdi\n" if ($narg>0); | |
346 | $func .= " movq %rdx,%rsi\n" if ($narg>1); | |
347 | $func .= " movq %r8,%rdx\n" if ($narg>2); | |
348 | $func .= " movq %r9,%rcx\n" if ($narg>3); | |
349 | $func .= " movq 40(%rsp),%r8\n" if ($narg>4); | |
350 | $func .= " movq 48(%rsp),%r9\n" if ($narg>5); | |
351 | } | |
352 | $func; | |
353 | } elsif ($self->{value} ne "$current_function->{name}") { | |
9b634c9b | 354 | $self->{value} .= ":" if ($masm && $ret!~m/^\$/); |
85253772 AP |
355 | $self->{value} . ":"; |
356 | } elsif ($win64 && $current_function->{abi} eq "svr4") { | |
9b634c9b AP |
357 | my $func = "$current_function->{name}" . |
358 | ($nasm ? ":" : "\tPROC $current_function->{scope}") . | |
359 | "\n"; | |
360 | $func .= " mov QWORD${PTR}[8+rsp],rdi\t;WIN64 prologue\n"; | |
361 | $func .= " mov QWORD${PTR}[16+rsp],rsi\n"; | |
80aa9cc9 AP |
362 | $func .= " mov rax,rsp\n"; |
363 | $func .= "\$L\$SEH_begin_$current_function->{name}:"; | |
364 | $func .= ":" if ($masm); | |
365 | $func .= "\n"; | |
1cfd258e AP |
366 | my $narg = $current_function->{narg}; |
367 | $narg=6 if (!defined($narg)); | |
368 | $func .= " mov rdi,rcx\n" if ($narg>0); | |
369 | $func .= " mov rsi,rdx\n" if ($narg>1); | |
370 | $func .= " mov rdx,r8\n" if ($narg>2); | |
371 | $func .= " mov rcx,r9\n" if ($narg>3); | |
a23e3dbe AP |
372 | $func .= " mov r8,QWORD${PTR}[40+rsp]\n" if ($narg>4); |
373 | $func .= " mov r9,QWORD${PTR}[48+rsp]\n" if ($narg>5); | |
1cfd258e AP |
374 | $func .= "\n"; |
375 | } else { | |
9b634c9b AP |
376 | "$current_function->{name}". |
377 | ($nasm ? ":" : "\tPROC $current_function->{scope}"); | |
1cfd258e AP |
378 | } |
379 | } | |
380 | } | |
381 | { package expr; # pick up expressioins | |
382 | sub re { | |
383 | my $self = shift; # single instance is enough... | |
384 | local *line = shift; | |
385 | undef $ret; | |
386 | ||
387 | if ($line =~ /(^[^,]+)/) { | |
388 | $self->{value} = $1; | |
389 | $ret = $self; | |
390 | $line = substr($line,@+[0]); $line =~ s/^\s+//; | |
391 | ||
85253772 AP |
392 | $self->{value} =~ s/\@PLT// if (!$elf); |
393 | $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; | |
394 | $self->{value} =~ s/\.L/\$L\$/g if (!$gas); | |
395 | $self->{value} =~ s/\.L/L\$/g if (!$elf); | |
1cfd258e AP |
396 | } |
397 | $ret; | |
398 | } | |
399 | sub out { | |
400 | my $self = shift; | |
9b634c9b AP |
401 | if ($nasm && opcode->mnemonic()=~m/^j/) { |
402 | "NEAR ".$self->{value}; | |
403 | } else { | |
404 | $self->{value}; | |
405 | } | |
1cfd258e AP |
406 | } |
407 | } | |
408 | { package directive; # pick up directives, which start with . | |
409 | sub re { | |
410 | my $self = shift; # single instance is enough... | |
411 | local *line = shift; | |
412 | undef $ret; | |
413 | my $dir; | |
5d0d60e2 AP |
414 | my %opcode = # lea 2f-1f(%rip),%dst; 1: nop; 2: |
415 | ( "%rax"=>0x01058d48, "%rcx"=>0x010d8d48, | |
416 | "%rdx"=>0x01158d48, "%rbx"=>0x011d8d48, | |
417 | "%rsp"=>0x01258d48, "%rbp"=>0x012d8d48, | |
418 | "%rsi"=>0x01358d48, "%rdi"=>0x013d8d48, | |
419 | "%r8" =>0x01058d4c, "%r9" =>0x010d8d4c, | |
420 | "%r10"=>0x01158d4c, "%r11"=>0x011d8d4c, | |
421 | "%r12"=>0x01258d4c, "%r13"=>0x012d8d4c, | |
422 | "%r14"=>0x01358d4c, "%r15"=>0x013d8d4c ); | |
1cfd258e AP |
423 | |
424 | if ($line =~ /^\s*(\.\w+)/) { | |
85253772 AP |
425 | $dir = $1; |
426 | $ret = $self; | |
427 | undef $self->{value}; | |
428 | $line = substr($line,@+[0]); $line =~ s/^\s+//; | |
429 | ||
430 | SWITCH: for ($dir) { | |
431 | /\.picmeup/ && do { if ($line =~ /(%r[\w]+)/i) { | |
432 | $dir="\t.long"; | |
433 | $line=sprintf "0x%x,0x90000000",$opcode{$1}; | |
434 | } | |
435 | last; | |
436 | }; | |
437 | /\.global|\.globl|\.extern/ | |
438 | && do { $globals{$line} = $prefix . $line; | |
439 | $line = $globals{$line} if ($prefix); | |
440 | last; | |
441 | }; | |
442 | /\.type/ && do { ($sym,$type,$narg) = split(',',$line); | |
443 | if ($type eq "\@function") { | |
444 | undef $current_function; | |
445 | $current_function->{name} = $sym; | |
446 | $current_function->{abi} = "svr4"; | |
447 | $current_function->{narg} = $narg; | |
448 | $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; | |
449 | } elsif ($type eq "\@abi-omnipotent") { | |
450 | undef $current_function; | |
451 | $current_function->{name} = $sym; | |
452 | $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; | |
453 | } | |
454 | $line =~ s/\@abi\-omnipotent/\@function/; | |
455 | $line =~ s/\@function.*/\@function/; | |
456 | last; | |
457 | }; | |
458 | /\.asciz/ && do { if ($line =~ /^"(.*)"$/) { | |
459 | $dir = ".byte"; | |
460 | $line = join(",",unpack("C*",$1),0); | |
461 | } | |
462 | last; | |
463 | }; | |
464 | } | |
465 | ||
466 | if ($gas) { | |
467 | $self->{value} = $dir . "\t" . $line; | |
468 | ||
469 | if ($dir =~ /\.extern/) { | |
932cc129 | 470 | $self->{value} = ""; # swallow extern |
85253772 AP |
471 | } elsif (!$elf && $dir =~ /\.type/) { |
472 | $self->{value} = ""; | |
473 | $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" . | |
474 | (defined($globals{$1})?".scl 2;":".scl 3;") . | |
475 | "\t.type 32;\t.endef" | |
476 | if ($win64 && $line =~ /([^,]+),\@function/); | |
477 | } elsif (!$elf && $dir =~ /\.size/) { | |
478 | $self->{value} = ""; | |
479 | if (defined($current_function)) { | |
480 | $self->{value} .= ".LSEH_end_$current_function->{name}:" | |
481 | if ($win64 && $current_function->{abi} eq "svr4"); | |
482 | undef $current_function; | |
483 | } | |
484 | } elsif (!$elf && $dir =~ /\.align/) { | |
485 | $self->{value} = ".p2align\t" . (log($line)/log(2)); | |
486 | } elsif ($dir eq ".section") { | |
487 | $current_segment=$line; | |
488 | if (!$elf && $current_segment eq ".init") { | |
489 | if ($flavour eq "macosx") { $self->{value} = ".mod_init_func"; } | |
490 | elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.ctors"; } | |
491 | } | |
492 | } elsif ($dir =~ /\.(text|data)/) { | |
493 | $current_segment=".$1"; | |
5d0d60e2 | 494 | } |
1cfd258e AP |
495 | $line = ""; |
496 | return $self; | |
497 | } | |
498 | ||
85253772 | 499 | # non-gas case or nasm/masm |
1cfd258e | 500 | SWITCH: for ($dir) { |
9b634c9b | 501 | /\.text/ && do { my $v=undef; |
a23e3dbe | 502 | if ($nasm) { |
9b634c9b | 503 | $v="section .text code align=64\n"; |
a23e3dbe AP |
504 | } else { |
505 | $v="$current_segment\tENDS\n" if ($current_segment); | |
9b634c9b | 506 | $current_segment = ".text\$"; |
a23e3dbe AP |
507 | $v.="$current_segment\tSEGMENT "; |
508 | $v.=$masm>=$masmref ? "ALIGN(64)" : "PAGE"; | |
509 | $v.=" 'CODE'"; | |
510 | } | |
1cfd258e AP |
511 | $self->{value} = $v; |
512 | last; | |
513 | }; | |
9b634c9b AP |
514 | /\.data/ && do { my $v=undef; |
515 | if ($nasm) { | |
516 | $v="section .data data align=8\n"; | |
517 | } else { | |
518 | $v="$current_segment\tENDS\n" if ($current_segment); | |
519 | $current_segment = "_DATA"; | |
520 | $v.="$current_segment\tSEGMENT"; | |
521 | } | |
522 | $self->{value} = $v; | |
523 | last; | |
524 | }; | |
525 | /\.section/ && do { my $v=undef; | |
85253772 AP |
526 | $line =~ s/([^,]*).*/$1/; |
527 | $line = ".CRT\$XCU" if ($line eq ".init"); | |
9b634c9b AP |
528 | if ($nasm) { |
529 | $v="section $line"; | |
530 | if ($line=~/\.([px])data/) { | |
531 | $v.=" rdata align="; | |
532 | $v.=$1 eq "p"? 4 : 8; | |
533 | } | |
534 | } else { | |
535 | $v="$current_segment\tENDS\n" if ($current_segment); | |
536 | $v.="$line\tSEGMENT"; | |
537 | if ($line=~/\.([px])data/) { | |
538 | $v.=" READONLY"; | |
539 | $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref); | |
540 | } | |
541 | } | |
542 | $current_segment = $line; | |
543 | $self->{value} = $v; | |
544 | last; | |
545 | }; | |
a23e3dbe | 546 | /\.extern/ && do { $self->{value} = "EXTERN\t".$line; |
9b634c9b AP |
547 | $self->{value} .= ":NEAR" if ($masm); |
548 | last; | |
549 | }; | |
80aa9cc9 AP |
550 | /\.globl|.global/ |
551 | && do { $self->{value} = $masm?"PUBLIC":"global"; | |
552 | $self->{value} .= "\t".$line; | |
1cfd258e AP |
553 | last; |
554 | }; | |
5d0d60e2 | 555 | /\.size/ && do { if (defined($current_function)) { |
8fe8bae1 AP |
556 | undef $self->{value}; |
557 | if ($current_function->{abi} eq "svr4") { | |
558 | $self->{value}="\$L\$SEH_end_$current_function->{name}:"; | |
559 | $self->{value}.=":\n" if($masm); | |
560 | } | |
561 | $self->{value}.="$current_function->{name}\tENDP" if($masm); | |
1cfd258e AP |
562 | undef $current_function; |
563 | } | |
564 | last; | |
565 | }; | |
566 | /\.align/ && do { $self->{value} = "ALIGN\t".$line; last; }; | |
85253772 | 567 | /\.(value|long|rva|quad)/ |
80aa9cc9 AP |
568 | && do { my $sz = substr($1,0,1); |
569 | my @arr = split(',',$line); | |
1cfd258e | 570 | my $last = pop(@arr); |
a23e3dbe | 571 | my $conv = sub { my $var=shift; |
9b634c9b | 572 | $var=~s/0x([0-9a-f]+)/0$1h/ig; |
80aa9cc9 | 573 | $var=~s/\.L/\$L\$/g; |
85253772 | 574 | if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva")) |
80aa9cc9 | 575 | { $var=~s/([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; } |
9b634c9b | 576 | $var; |
a23e3dbe | 577 | }; |
1cfd258e | 578 | |
85253772 | 579 | $sz =~ tr/bvlrq/BWDDQ/; |
1cfd258e | 580 | $self->{value} = "\tD$sz\t"; |
a23e3dbe AP |
581 | for (@arr) { $self->{value} .= &$conv($_).","; } |
582 | $self->{value} .= &$conv($last); | |
1cfd258e AP |
583 | last; |
584 | }; | |
85253772 AP |
585 | /\.byte/ && do { my @str=split(",",$line); |
586 | while ($#str>15) { | |
55eab3b7 | 587 | $self->{value}.="DB\t" |
85253772 AP |
588 | .join(",",@str[0..15])."\n"; |
589 | foreach (0..15) { shift @str; } | |
d68ff710 | 590 | } |
85253772 AP |
591 | $self->{value}.="DB\t" |
592 | .join(",",@str) if (@str); | |
d68ff710 AP |
593 | last; |
594 | }; | |
1cfd258e AP |
595 | } |
596 | $line = ""; | |
597 | } | |
598 | ||
599 | $ret; | |
600 | } | |
601 | sub out { | |
602 | my $self = shift; | |
603 | $self->{value}; | |
604 | } | |
605 | } | |
606 | ||
9b634c9b AP |
607 | if ($nasm) { |
608 | print <<___; | |
609 | default rel | |
9b634c9b AP |
610 | ___ |
611 | } elsif ($masm) { | |
612 | print <<___; | |
613 | OPTION DOTNAME | |
614 | ___ | |
615 | } | |
1cfd258e AP |
616 | while($line=<>) { |
617 | ||
618 | chomp($line); | |
619 | ||
4b450519 AP |
620 | $line =~ s|[#!].*$||; # get rid of asm-style comments... |
621 | $line =~ s|/\*.*\*/||; # ... and C-style comments... | |
622 | $line =~ s|^\s+||; # ... and skip white spaces in beginning | |
1cfd258e AP |
623 | |
624 | undef $label; | |
625 | undef $opcode; | |
626 | undef $dst; | |
627 | undef $src; | |
628 | undef $sz; | |
629 | ||
630 | if ($label=label->re(\$line)) { print $label->out(); } | |
631 | ||
632 | if (directive->re(\$line)) { | |
633 | printf "%s",directive->out(); | |
634 | } elsif ($opcode=opcode->re(\$line)) { ARGUMENT: { | |
635 | ||
636 | if ($src=register->re(\$line)) { opcode->size($src->size()); } | |
637 | elsif ($src=const->re(\$line)) { } | |
638 | elsif ($src=ea->re(\$line)) { } | |
639 | elsif ($src=expr->re(\$line)) { } | |
640 | ||
641 | last ARGUMENT if ($line !~ /^,/); | |
642 | ||
643 | $line = substr($line,1); $line =~ s/^\s+//; | |
644 | ||
645 | if ($dst=register->re(\$line)) { opcode->size($dst->size()); } | |
646 | elsif ($dst=const->re(\$line)) { } | |
647 | elsif ($dst=ea->re(\$line)) { } | |
648 | ||
649 | } # ARGUMENT: | |
650 | ||
651 | $sz=opcode->size(); | |
652 | ||
653 | if (defined($dst)) { | |
85253772 | 654 | if ($gas) { |
1cfd258e AP |
655 | printf "\t%s\t%s,%s", $opcode->out($dst->size()), |
656 | $src->out($sz),$dst->out($sz); | |
5d0d60e2 | 657 | } else { |
a23e3dbe | 658 | undef $sz if ($nasm && $opcode->mnemonic() eq "lea"); |
1cfd258e AP |
659 | printf "\t%s\t%s,%s", $opcode->out(), |
660 | $dst->out($sz),$src->out($sz); | |
661 | } | |
5d0d60e2 | 662 | } elsif (defined($src)) { |
1cfd258e AP |
663 | printf "\t%s\t%s",$opcode->out(),$src->out($sz); |
664 | } else { | |
665 | printf "\t%s",$opcode->out(); | |
666 | } | |
667 | } | |
668 | ||
669 | print $line,"\n"; | |
670 | } | |
671 | ||
85253772 AP |
672 | print "\n$current_segment\tENDS\n" if ($current_segment && $masm); |
673 | print "END\n" if ($masm); | |
1cfd258e AP |
674 | |
675 | close STDOUT; | |
676 | ||
80aa9cc9 | 677 | \f################################################# |
1cfd258e AP |
678 | # Cross-reference x86_64 ABI "card" |
679 | # | |
680 | # Unix Win64 | |
681 | # %rax * * | |
682 | # %rbx - - | |
683 | # %rcx #4 #1 | |
684 | # %rdx #3 #2 | |
685 | # %rsi #2 - | |
686 | # %rdi #1 - | |
687 | # %rbp - - | |
688 | # %rsp - - | |
689 | # %r8 #5 #3 | |
690 | # %r9 #6 #4 | |
691 | # %r10 * * | |
692 | # %r11 * * | |
693 | # %r12 - - | |
694 | # %r13 - - | |
695 | # %r14 - - | |
696 | # %r15 - - | |
697 | # | |
698 | # (*) volatile register | |
699 | # (-) preserved by callee | |
700 | # (#) Nth argument, volatile | |
701 | # | |
702 | # In Unix terms top of stack is argument transfer area for arguments | |
703 | # which could not be accomodated in registers. Or in other words 7th | |
704 | # [integer] argument resides at 8(%rsp) upon function entry point. | |
705 | # 128 bytes above %rsp constitute a "red zone" which is not touched | |
706 | # by signal handlers and can be used as temporal storage without | |
707 | # allocating a frame. | |
708 | # | |
709 | # In Win64 terms N*8 bytes on top of stack is argument transfer area, | |
710 | # which belongs to/can be overwritten by callee. N is the number of | |
711 | # arguments passed to callee, *but* not less than 4! This means that | |
712 | # upon function entry point 5th argument resides at 40(%rsp), as well | |
713 | # as that 32 bytes from 8(%rsp) can always be used as temporal | |
d256b957 AP |
714 | # storage [without allocating a frame]. One can actually argue that |
715 | # one can assume a "red zone" above stack pointer under Win64 as well. | |
6a3a7f30 AP |
716 | # Point is that at apparently no occasion Windows kernel would alter |
717 | # the area above user stack pointer in true asynchronous manner... | |
1cfd258e AP |
718 | # |
719 | # All the above means that if assembler programmer adheres to Unix | |
720 | # register and stack layout, but disregards the "red zone" existense, | |
721 | # it's possible to use following prologue and epilogue to "gear" from | |
722 | # Unix to Win64 ABI in leaf functions with not more than 6 arguments. | |
723 | # | |
724 | # omnipotent_function: | |
725 | # ifdef WIN64 | |
726 | # movq %rdi,8(%rsp) | |
727 | # movq %rsi,16(%rsp) | |
728 | # movq %rcx,%rdi ; if 1st argument is actually present | |
729 | # movq %rdx,%rsi ; if 2nd argument is actually ... | |
730 | # movq %r8,%rdx ; if 3rd argument is ... | |
731 | # movq %r9,%rcx ; if 4th argument ... | |
732 | # movq 40(%rsp),%r8 ; if 5th ... | |
733 | # movq 48(%rsp),%r9 ; if 6th ... | |
734 | # endif | |
735 | # ... | |
736 | # ifdef WIN64 | |
737 | # movq 8(%rsp),%rdi | |
738 | # movq 16(%rsp),%rsi | |
739 | # endif | |
740 | # ret | |
4f469342 | 741 | # |
80aa9cc9 AP |
742 | \f################################################# |
743 | # Win64 SEH, Structured Exception Handling. | |
744 | # | |
4f469342 AP |
745 | # Unlike on Unix systems(*) lack of Win64 stack unwinding information |
746 | # has undesired side-effect at run-time: if an exception is raised in | |
747 | # assembler subroutine such as those in question (basically we're | |
748 | # referring to segmentation violations caused by malformed input | |
749 | # parameters), the application is briskly terminated without invoking | |
750 | # any exception handlers, most notably without generating memory dump | |
751 | # or any user notification whatsoever. This poses a problem. It's | |
752 | # possible to address it by registering custom language-specific | |
753 | # handler that would restore processor context to the state at | |
754 | # subroutine entry point and return "exception is not handled, keep | |
755 | # unwinding" code. Writing such handler can be a challenge... But it's | |
756 | # doable, though requires certain coding convention. Consider following | |
757 | # snippet: | |
758 | # | |
80aa9cc9 | 759 | # .type function,@function |
4f469342 AP |
760 | # function: |
761 | # movq %rsp,%rax # copy rsp to volatile register | |
762 | # pushq %r15 # save non-volatile registers | |
763 | # pushq %rbx | |
764 | # pushq %rbp | |
765 | # movq %rsp,%r11 | |
766 | # subq %rdi,%r11 # prepare [variable] stack frame | |
767 | # andq $-64,%r11 | |
768 | # movq %rax,0(%r11) # check for exceptions | |
769 | # movq %r11,%rsp # allocate [variable] stack frame | |
770 | # movq %rax,0(%rsp) # save original rsp value | |
771 | # magic_point: | |
772 | # ... | |
773 | # movq 0(%rsp),%rcx # pull original rsp value | |
774 | # movq -24(%rcx),%rbp # restore non-volatile registers | |
775 | # movq -16(%rcx),%rbx | |
776 | # movq -8(%rcx),%r15 | |
777 | # movq %rcx,%rsp # restore original rsp | |
778 | # ret | |
80aa9cc9 | 779 | # .size function,.-function |
4f469342 AP |
780 | # |
781 | # The key is that up to magic_point copy of original rsp value remains | |
782 | # in chosen volatile register and no non-volatile register, except for | |
783 | # rsp, is modified. While past magic_point rsp remains constant till | |
784 | # the very end of the function. In this case custom language-specific | |
785 | # exception handler would look like this: | |
786 | # | |
787 | # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, | |
788 | # CONTEXT *context,DISPATCHER_CONTEXT *disp) | |
80aa9cc9 AP |
789 | # { ULONG64 *rsp = (ULONG64 *)context->Rax; |
790 | # if (context->Rip >= magic_point) | |
4f469342 AP |
791 | # { rsp = ((ULONG64 **)context->Rsp)[0]; |
792 | # context->Rbp = rsp[-3]; | |
793 | # context->Rbx = rsp[-2]; | |
794 | # context->R15 = rsp[-1]; | |
795 | # } | |
796 | # context->Rsp = (ULONG64)rsp; | |
797 | # context->Rdi = rsp[1]; | |
798 | # context->Rsi = rsp[2]; | |
799 | # | |
800 | # memcpy (disp->ContextRecord,context,sizeof(CONTEXT)); | |
801 | # RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase, | |
802 | # dips->ControlPc,disp->FunctionEntry,disp->ContextRecord, | |
803 | # &disp->HandlerData,&disp->EstablisherFrame,NULL); | |
804 | # return ExceptionContinueSearch; | |
805 | # } | |
806 | # | |
807 | # It's appropriate to implement this handler in assembler, directly in | |
808 | # function's module. In order to do that one has to know members' | |
809 | # offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant | |
810 | # values. Here they are: | |
811 | # | |
812 | # CONTEXT.Rax 120 | |
813 | # CONTEXT.Rcx 128 | |
814 | # CONTEXT.Rdx 136 | |
815 | # CONTEXT.Rbx 144 | |
816 | # CONTEXT.Rsp 152 | |
817 | # CONTEXT.Rbp 160 | |
818 | # CONTEXT.Rsi 168 | |
819 | # CONTEXT.Rdi 176 | |
820 | # CONTEXT.R8 184 | |
821 | # CONTEXT.R9 192 | |
822 | # CONTEXT.R10 200 | |
823 | # CONTEXT.R11 208 | |
824 | # CONTEXT.R12 216 | |
825 | # CONTEXT.R13 224 | |
826 | # CONTEXT.R14 232 | |
827 | # CONTEXT.R15 240 | |
828 | # CONTEXT.Rip 248 | |
829 | # sizeof(CONTEXT) 1232 | |
830 | # DISPATCHER_CONTEXT.ControlPc 0 | |
831 | # DISPATCHER_CONTEXT.ImageBase 8 | |
832 | # DISPATCHER_CONTEXT.FunctionEntry 16 | |
833 | # DISPATCHER_CONTEXT.EstablisherFrame 24 | |
834 | # DISPATCHER_CONTEXT.TargetIp 32 | |
835 | # DISPATCHER_CONTEXT.ContextRecord 40 | |
836 | # DISPATCHER_CONTEXT.LanguageHandler 48 | |
837 | # DISPATCHER_CONTEXT.HandlerData 56 | |
838 | # UNW_FLAG_NHANDLER 0 | |
839 | # ExceptionContinueSearch 1 | |
840 | # | |
80aa9cc9 AP |
841 | # In order to tie the handler to the function one has to compose |
842 | # couple of structures: one for .xdata segment and one for .pdata. | |
843 | # | |
4f469342 | 844 | # UNWIND_INFO structure for .xdata segment would be |
80aa9cc9 AP |
845 | # |
846 | # function_unwind_info: | |
847 | # .byte 9,0,0,0 | |
85253772 | 848 | # .rva handler |
80aa9cc9 AP |
849 | # |
850 | # This structure designates exception handler for a function with | |
851 | # zero-length prologue, no stack frame or frame register. | |
852 | # | |
853 | # To facilitate composing of .pdata structures, auto-generated "gear" | |
854 | # prologue copies rsp value to rax and denotes next instruction with | |
85253772 | 855 | # .LSEH_begin_{function_name} label. This essentially defines the SEH |
80aa9cc9 AP |
856 | # styling rule mentioned in the beginning. Position of this label is |
857 | # chosen in such manner that possible exceptions raised in the "gear" | |
858 | # prologue would be accounted to caller and unwound from latter's frame. | |
85253772 | 859 | # End of function is marked with respective .LSEH_end_{function_name} |
80aa9cc9 AP |
860 | # label. To summarize, .pdata segment would contain |
861 | # | |
85253772 AP |
862 | # .rva .LSEH_begin_function |
863 | # .rva .LSEH_end_function | |
864 | # .rva function_unwind_info | |
80aa9cc9 AP |
865 | # |
866 | # Reference to functon_unwind_info from .xdata segment is the anchor. | |
85253772 | 867 | # In case you wonder why references are 32-bit .rvas and not 64-bit |
80aa9cc9 AP |
868 | # .quads. References put into these two segments are required to be |
869 | # *relative* to the base address of the current binary module, a.k.a. | |
870 | # image base. No Win64 module, be it .exe or .dll, can be larger than | |
871 | # 2GB and thus such relative references can be and are accommodated in | |
872 | # 32 bits. | |
873 | # | |
874 | # Having reviewed the example function code, one can argue that "movq | |
875 | # %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix | |
876 | # rax would contain an undefined value. If this "offends" you, use | |
877 | # another register and refrain from modifying rax till magic_point is | |
878 | # reached, i.e. as if it was a non-volatile register. If more registers | |
879 | # are required prior [variable] frame setup is completed, note that | |
880 | # nobody says that you can have only one "magic point." You can | |
881 | # "liberate" non-volatile registers by denoting last stack off-load | |
882 | # instruction and reflecting it in finer grade unwind logic in handler. | |
883 | # After all, isn't it why it's called *language-specific* handler... | |
884 | # | |
885 | # Attentive reader can notice that exceptions would be mishandled in | |
886 | # auto-generated "gear" epilogue. Well, exception effectively can't | |
887 | # occur there, because if memory area used by it was subject to | |
888 | # segmentation violation, then it would be raised upon call to the | |
889 | # function (and as already mentioned be accounted to caller, which is | |
890 | # not a problem). If you're still not comfortable, then define tail | |
891 | # "magic point" just prior ret instruction and have handler treat it... | |
4f469342 AP |
892 | # |
893 | # (*) Note that we're talking about run-time, not debug-time. Lack of | |
894 | # unwind information makes debugging hard on both Windows and | |
895 | # Unix. "Unlike" referes to the fact that on Unix signal handler | |
896 | # will always be invoked, core dumped and appropriate exit code | |
897 | # returned to parent (for user notification). |