]>
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. |
f9a152bd AP |
58 | # a. If function accepts more than 4 arguments *and* >4th argument |
59 | # is declared as non 64-bit value, do clear its upper part. | |
80aa9cc9 | 60 | \f |
85253772 AP |
61 | my $flavour = shift; |
62 | my $output = shift; | |
63 | if ($flavour =~ /\./) { $output = $flavour; undef $flavour; } | |
932cc129 | 64 | |
b2ae61ec AP |
65 | open STDOUT,">$output" || die "can't open $output: $!" |
66 | if (defined($output)); | |
1cfd258e | 67 | |
85253772 AP |
68 | my $gas=1; $gas=0 if ($output =~ /\.asm$/); |
69 | my $elf=1; $elf=0 if (!$gas); | |
70 | my $win64=0; | |
71 | my $prefix=""; | |
93c4ba07 | 72 | my $decor=".L"; |
a23e3dbe | 73 | |
8ab9025e | 74 | my $masmref=8 + 50727*2**-32; # 8.00.50727 shipped with VS2005 |
9b634c9b | 75 | my $masm=0; |
a23e3dbe AP |
76 | my $PTR=" PTR"; |
77 | ||
9b634c9b | 78 | my $nasmref=2.03; |
a23e3dbe AP |
79 | my $nasm=0; |
80 | ||
f9a152bd AP |
81 | if ($flavour eq "mingw64") { $gas=1; $elf=0; $win64=1; |
82 | $prefix=`echo __USER_LABEL_PREFIX__ | $ENV{CC} -E -P -`; | |
83 | chomp($prefix); | |
84 | } | |
93c4ba07 AP |
85 | elsif ($flavour eq "macosx") { $gas=1; $elf=0; $prefix="_"; $decor="L\$"; } |
86 | elsif ($flavour eq "masm") { $gas=0; $elf=0; $masm=$masmref; $win64=1; $decor="\$L\$"; } | |
87 | elsif ($flavour eq "nasm") { $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $decor="\$L\$"; $PTR=""; } | |
85253772 | 88 | elsif (!$gas) |
9b634c9b AP |
89 | { if ($ENV{ASM} =~ m/nasm/ && `nasm -v` =~ m/version ([0-9]+)\.([0-9]+)/i) |
90 | { $nasm = $1 + $2*0.01; $PTR=""; } | |
a23e3dbe AP |
91 | elsif (`ml64 2>&1` =~ m/Version ([0-9]+)\.([0-9]+)(\.([0-9]+))?/) |
92 | { $masm = $1 + $2*2**-16 + $4*2**-32; } | |
9b634c9b | 93 | die "no assembler found on %PATH" if (!($nasm || $masm)); |
85253772 AP |
94 | $win64=1; |
95 | $elf=0; | |
93c4ba07 | 96 | $decor="\$L\$"; |
a23e3dbe | 97 | } |
1cfd258e AP |
98 | |
99 | my $current_segment; | |
100 | my $current_function; | |
9b634c9b | 101 | my %globals; |
1cfd258e AP |
102 | |
103 | { package opcode; # pick up opcodes | |
104 | sub re { | |
105 | my $self = shift; # single instance in enough... | |
106 | local *line = shift; | |
107 | undef $ret; | |
108 | ||
932cc129 | 109 | if ($line =~ /^([a-z][a-z0-9]*)/i) { |
1cfd258e AP |
110 | $self->{op} = $1; |
111 | $ret = $self; | |
112 | $line = substr($line,@+[0]); $line =~ s/^\s+//; | |
113 | ||
114 | undef $self->{sz}; | |
272ba870 | 115 | if ($self->{op} =~ /^(movz)x?([bw]).*/) { # movz is pain... |
1cfd258e | 116 | $self->{op} = $1; |
272ba870 | 117 | $self->{sz} = $2; |
9b634c9b | 118 | } elsif ($self->{op} =~ /call|jmp/) { |
fead2539 | 119 | $self->{sz} = ""; |
a9e790b9 | 120 | } elsif ($self->{op} =~ /^p/ && $' !~ /^(ush|op|insrw)/) { # SSEn |
fead2539 | 121 | $self->{sz} = ""; |
0a9a692e AP |
122 | } elsif ($self->{op} =~ /^v/) { # VEX |
123 | $self->{sz} = ""; | |
c7b903e0 AP |
124 | } elsif ($self->{op} =~ /movq/ && $line =~ /%xmm/) { |
125 | $self->{sz} = ""; | |
a078befc | 126 | } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) { |
1cfd258e AP |
127 | $self->{op} = $1; |
128 | $self->{sz} = $2; | |
129 | } | |
130 | } | |
131 | $ret; | |
132 | } | |
133 | sub size { | |
134 | my $self = shift; | |
135 | my $sz = shift; | |
136 | $self->{sz} = $sz if (defined($sz) && !defined($self->{sz})); | |
137 | $self->{sz}; | |
138 | } | |
139 | sub out { | |
140 | my $self = shift; | |
85253772 | 141 | if ($gas) { |
932cc129 | 142 | if ($self->{op} eq "movz") { # movz is pain... |
1cfd258e | 143 | sprintf "%s%s%s",$self->{op},$self->{sz},shift; |
932cc129 AP |
144 | } elsif ($self->{op} =~ /^set/) { |
145 | "$self->{op}"; | |
1cfd258e | 146 | } elsif ($self->{op} eq "ret") { |
85253772 AP |
147 | my $epilogue = ""; |
148 | if ($win64 && $current_function->{abi} eq "svr4") { | |
149 | $epilogue = "movq 8(%rsp),%rdi\n\t" . | |
150 | "movq 16(%rsp),%rsi\n\t"; | |
151 | } | |
152 | $epilogue . ".byte 0xf3,0xc3"; | |
153 | } elsif ($self->{op} eq "call" && !$elf && $current_segment eq ".init") { | |
154 | ".p2align\t3\n\t.quad"; | |
1cfd258e AP |
155 | } else { |
156 | "$self->{op}$self->{sz}"; | |
157 | } | |
158 | } else { | |
a078befc | 159 | $self->{op} =~ s/^movz/movzx/; |
1cfd258e AP |
160 | if ($self->{op} eq "ret") { |
161 | $self->{op} = ""; | |
85253772 | 162 | if ($win64 && $current_function->{abi} eq "svr4") { |
a23e3dbe AP |
163 | $self->{op} = "mov rdi,QWORD${PTR}[8+rsp]\t;WIN64 epilogue\n\t". |
164 | "mov rsi,QWORD${PTR}[16+rsp]\n\t"; | |
1cfd258e AP |
165 | } |
166 | $self->{op} .= "DB\t0F3h,0C3h\t\t;repret"; | |
85253772 | 167 | } elsif ($self->{op} =~ /^(pop|push)f/) { |
80aa9cc9 | 168 | $self->{op} .= $self->{sz}; |
85253772 | 169 | } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") { |
d6522548 | 170 | $self->{op} = "\tDQ"; |
85253772 | 171 | } |
1cfd258e AP |
172 | $self->{op}; |
173 | } | |
174 | } | |
85253772 AP |
175 | sub mnemonic { |
176 | my $self=shift; | |
177 | my $op=shift; | |
178 | $self->{op}=$op if (defined($op)); | |
179 | $self->{op}; | |
180 | } | |
1cfd258e AP |
181 | } |
182 | { package const; # pick up constants, which start with $ | |
183 | sub re { | |
184 | my $self = shift; # single instance in enough... | |
185 | local *line = shift; | |
186 | undef $ret; | |
187 | ||
188 | if ($line =~ /^\$([^,]+)/) { | |
189 | $self->{value} = $1; | |
190 | $ret = $self; | |
191 | $line = substr($line,@+[0]); $line =~ s/^\s+//; | |
192 | } | |
193 | $ret; | |
194 | } | |
195 | sub out { | |
196 | my $self = shift; | |
5d0d60e2 | 197 | |
85253772 | 198 | if ($gas) { |
e84b663a AP |
199 | # Solaris /usr/ccs/bin/as can't handle multiplications |
200 | # in $self->{value} | |
f9a152bd | 201 | $self->{value} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi; |
e84b663a | 202 | $self->{value} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg; |
5d0d60e2 AP |
203 | sprintf "\$%s",$self->{value}; |
204 | } else { | |
e8169520 AP |
205 | $self->{value} =~ s/(0b[0-1]+)/oct($1)/eig; |
206 | $self->{value} =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm); | |
5d0d60e2 AP |
207 | sprintf "%s",$self->{value}; |
208 | } | |
1cfd258e AP |
209 | } |
210 | } | |
211 | { package ea; # pick up effective addresses: expr(%reg,%reg,scale) | |
212 | sub re { | |
213 | my $self = shift; # single instance in enough... | |
214 | local *line = shift; | |
215 | undef $ret; | |
216 | ||
9b634c9b AP |
217 | # optional * ---vvv--- appears in indirect jmp/call |
218 | if ($line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)/) { | |
219 | $self->{asterisk} = $1; | |
220 | $self->{label} = $2; | |
221 | ($self->{base},$self->{index},$self->{scale})=split(/,/,$3); | |
1cfd258e AP |
222 | $self->{scale} = 1 if (!defined($self->{scale})); |
223 | $ret = $self; | |
224 | $line = substr($line,@+[0]); $line =~ s/^\s+//; | |
225 | ||
85253772 AP |
226 | if ($win64 && $self->{label} =~ s/\@GOTPCREL//) { |
227 | die if (opcode->mnemonic() ne "mov"); | |
228 | opcode->mnemonic("lea"); | |
229 | } | |
1cfd258e AP |
230 | $self->{base} =~ s/^%//; |
231 | $self->{index} =~ s/^%// if (defined($self->{index})); | |
232 | } | |
233 | $ret; | |
234 | } | |
235 | sub size {} | |
236 | sub out { | |
237 | my $self = shift; | |
238 | my $sz = shift; | |
239 | ||
85253772 | 240 | $self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; |
93c4ba07 | 241 | $self->{label} =~ s/\.L/$decor/g; |
85253772 | 242 | |
afbe674e AP |
243 | # Silently convert all EAs to 64-bit. This is required for |
244 | # elder GNU assembler and results in more compact code, | |
245 | # *but* most importantly AES module depends on this feature! | |
246 | $self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; | |
247 | $self->{base} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/; | |
248 | ||
6fa4c7c4 AP |
249 | # Solaris /usr/ccs/bin/as can't handle multiplications |
250 | # in $self->{label}, new gas requires sign extension... | |
251 | use integer; | |
252 | $self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi; | |
253 | $self->{label} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg; | |
254 | $self->{label} =~ s/([0-9]+)/$1<<32>>32/eg; | |
255 | ||
85253772 | 256 | if ($gas) { |
93c4ba07 | 257 | $self->{label} =~ s/^___imp_/__imp__/ if ($flavour eq "mingw64"); |
4b450519 | 258 | |
1cfd258e | 259 | if (defined($self->{index})) { |
b5c6aab5 AP |
260 | sprintf "%s%s(%s,%%%s,%d)",$self->{asterisk}, |
261 | $self->{label}, | |
262 | $self->{base}?"%$self->{base}":"", | |
1cfd258e | 263 | $self->{index},$self->{scale}; |
5d0d60e2 | 264 | } else { |
9b634c9b | 265 | sprintf "%s%s(%%%s)", $self->{asterisk},$self->{label},$self->{base}; |
1cfd258e AP |
266 | } |
267 | } else { | |
046ea308 | 268 | %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR", l=>"DWORD$PTR", |
a9e790b9 | 269 | q=>"QWORD$PTR",o=>"OWORD$PTR",x=>"XMMWORD$PTR" ); |
1cfd258e | 270 | |
70cf3095 | 271 | $self->{label} =~ s/\./\$/g; |
f9a152bd | 272 | $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig; |
70cf3095 | 273 | $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/); |
94c64f9a | 274 | $sz="q" if ($self->{asterisk} || opcode->mnemonic() eq "movq"); |
4010b341 | 275 | $sz="l" if (opcode->mnemonic() eq "movd"); |
70cf3095 | 276 | |
1cfd258e | 277 | if (defined($self->{index})) { |
b5c6aab5 | 278 | sprintf "%s[%s%s*%d%s]",$szmap{$sz}, |
a23e3dbe | 279 | $self->{label}?"$self->{label}+":"", |
1cfd258e | 280 | $self->{index},$self->{scale}, |
b5c6aab5 | 281 | $self->{base}?"+$self->{base}":""; |
932cc129 | 282 | } elsif ($self->{base} eq "rip") { |
a23e3dbe | 283 | sprintf "%s[%s]",$szmap{$sz},$self->{label}; |
5d0d60e2 | 284 | } else { |
a23e3dbe AP |
285 | sprintf "%s[%s%s]",$szmap{$sz}, |
286 | $self->{label}?"$self->{label}+":"", | |
287 | $self->{base}; | |
1cfd258e AP |
288 | } |
289 | } | |
290 | } | |
291 | } | |
292 | { package register; # pick up registers, which start with %. | |
293 | sub re { | |
294 | my $class = shift; # muliple instances... | |
295 | my $self = {}; | |
296 | local *line = shift; | |
297 | undef $ret; | |
298 | ||
9b634c9b AP |
299 | # optional * ---vvv--- appears in indirect jmp/call |
300 | if ($line =~ /^(\*?)%(\w+)/) { | |
1cfd258e | 301 | bless $self,$class; |
9b634c9b AP |
302 | $self->{asterisk} = $1; |
303 | $self->{value} = $2; | |
1cfd258e AP |
304 | $ret = $self; |
305 | $line = substr($line,@+[0]); $line =~ s/^\s+//; | |
306 | } | |
307 | $ret; | |
308 | } | |
309 | sub size { | |
310 | my $self = shift; | |
311 | undef $ret; | |
312 | ||
313 | if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; } | |
314 | elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; } | |
315 | elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; } | |
316 | elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; } | |
317 | elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; } | |
318 | elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; } | |
319 | elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; } | |
320 | elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; } | |
321 | ||
322 | $ret; | |
323 | } | |
324 | sub out { | |
325 | my $self = shift; | |
85253772 | 326 | if ($gas) { sprintf "%s%%%s",$self->{asterisk},$self->{value}; } |
9b634c9b | 327 | else { $self->{value}; } |
1cfd258e AP |
328 | } |
329 | } | |
330 | { package label; # pick up labels, which end with : | |
331 | sub re { | |
332 | my $self = shift; # single instance is enough... | |
333 | local *line = shift; | |
334 | undef $ret; | |
335 | ||
85253772 | 336 | if ($line =~ /(^[\.\w]+)\:/) { |
1cfd258e AP |
337 | $self->{value} = $1; |
338 | $ret = $self; | |
339 | $line = substr($line,@+[0]); $line =~ s/^\s+//; | |
340 | ||
93c4ba07 | 341 | $self->{value} =~ s/^\.L/$decor/; |
1cfd258e AP |
342 | } |
343 | $ret; | |
344 | } | |
345 | sub out { | |
346 | my $self = shift; | |
347 | ||
85253772 AP |
348 | if ($gas) { |
349 | my $func = ($globals{$self->{value}} or $self->{value}) . ":"; | |
350 | if ($win64 && | |
351 | $current_function->{name} eq $self->{value} && | |
352 | $current_function->{abi} eq "svr4") { | |
353 | $func .= "\n"; | |
354 | $func .= " movq %rdi,8(%rsp)\n"; | |
355 | $func .= " movq %rsi,16(%rsp)\n"; | |
356 | $func .= " movq %rsp,%rax\n"; | |
93c4ba07 | 357 | $func .= "${decor}SEH_begin_$current_function->{name}:\n"; |
85253772 AP |
358 | my $narg = $current_function->{narg}; |
359 | $narg=6 if (!defined($narg)); | |
360 | $func .= " movq %rcx,%rdi\n" if ($narg>0); | |
361 | $func .= " movq %rdx,%rsi\n" if ($narg>1); | |
362 | $func .= " movq %r8,%rdx\n" if ($narg>2); | |
363 | $func .= " movq %r9,%rcx\n" if ($narg>3); | |
364 | $func .= " movq 40(%rsp),%r8\n" if ($narg>4); | |
365 | $func .= " movq 48(%rsp),%r9\n" if ($narg>5); | |
366 | } | |
367 | $func; | |
368 | } elsif ($self->{value} ne "$current_function->{name}") { | |
9b634c9b | 369 | $self->{value} .= ":" if ($masm && $ret!~m/^\$/); |
85253772 AP |
370 | $self->{value} . ":"; |
371 | } elsif ($win64 && $current_function->{abi} eq "svr4") { | |
9b634c9b AP |
372 | my $func = "$current_function->{name}" . |
373 | ($nasm ? ":" : "\tPROC $current_function->{scope}") . | |
374 | "\n"; | |
375 | $func .= " mov QWORD${PTR}[8+rsp],rdi\t;WIN64 prologue\n"; | |
376 | $func .= " mov QWORD${PTR}[16+rsp],rsi\n"; | |
80aa9cc9 | 377 | $func .= " mov rax,rsp\n"; |
93c4ba07 | 378 | $func .= "${decor}SEH_begin_$current_function->{name}:"; |
80aa9cc9 AP |
379 | $func .= ":" if ($masm); |
380 | $func .= "\n"; | |
1cfd258e AP |
381 | my $narg = $current_function->{narg}; |
382 | $narg=6 if (!defined($narg)); | |
383 | $func .= " mov rdi,rcx\n" if ($narg>0); | |
384 | $func .= " mov rsi,rdx\n" if ($narg>1); | |
385 | $func .= " mov rdx,r8\n" if ($narg>2); | |
386 | $func .= " mov rcx,r9\n" if ($narg>3); | |
a23e3dbe AP |
387 | $func .= " mov r8,QWORD${PTR}[40+rsp]\n" if ($narg>4); |
388 | $func .= " mov r9,QWORD${PTR}[48+rsp]\n" if ($narg>5); | |
1cfd258e AP |
389 | $func .= "\n"; |
390 | } else { | |
9b634c9b AP |
391 | "$current_function->{name}". |
392 | ($nasm ? ":" : "\tPROC $current_function->{scope}"); | |
1cfd258e AP |
393 | } |
394 | } | |
395 | } | |
396 | { package expr; # pick up expressioins | |
397 | sub re { | |
398 | my $self = shift; # single instance is enough... | |
399 | local *line = shift; | |
400 | undef $ret; | |
401 | ||
402 | if ($line =~ /(^[^,]+)/) { | |
403 | $self->{value} = $1; | |
404 | $ret = $self; | |
405 | $line = substr($line,@+[0]); $line =~ s/^\s+//; | |
406 | ||
85253772 AP |
407 | $self->{value} =~ s/\@PLT// if (!$elf); |
408 | $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; | |
93c4ba07 | 409 | $self->{value} =~ s/\.L/$decor/g; |
1cfd258e AP |
410 | } |
411 | $ret; | |
412 | } | |
413 | sub out { | |
414 | my $self = shift; | |
9b634c9b AP |
415 | if ($nasm && opcode->mnemonic()=~m/^j/) { |
416 | "NEAR ".$self->{value}; | |
417 | } else { | |
418 | $self->{value}; | |
419 | } | |
1cfd258e AP |
420 | } |
421 | } | |
422 | { package directive; # pick up directives, which start with . | |
423 | sub re { | |
424 | my $self = shift; # single instance is enough... | |
425 | local *line = shift; | |
426 | undef $ret; | |
427 | my $dir; | |
5d0d60e2 AP |
428 | my %opcode = # lea 2f-1f(%rip),%dst; 1: nop; 2: |
429 | ( "%rax"=>0x01058d48, "%rcx"=>0x010d8d48, | |
430 | "%rdx"=>0x01158d48, "%rbx"=>0x011d8d48, | |
431 | "%rsp"=>0x01258d48, "%rbp"=>0x012d8d48, | |
432 | "%rsi"=>0x01358d48, "%rdi"=>0x013d8d48, | |
433 | "%r8" =>0x01058d4c, "%r9" =>0x010d8d4c, | |
434 | "%r10"=>0x01158d4c, "%r11"=>0x011d8d4c, | |
435 | "%r12"=>0x01258d4c, "%r13"=>0x012d8d4c, | |
436 | "%r14"=>0x01358d4c, "%r15"=>0x013d8d4c ); | |
1cfd258e AP |
437 | |
438 | if ($line =~ /^\s*(\.\w+)/) { | |
85253772 AP |
439 | $dir = $1; |
440 | $ret = $self; | |
441 | undef $self->{value}; | |
442 | $line = substr($line,@+[0]); $line =~ s/^\s+//; | |
443 | ||
444 | SWITCH: for ($dir) { | |
445 | /\.picmeup/ && do { if ($line =~ /(%r[\w]+)/i) { | |
446 | $dir="\t.long"; | |
447 | $line=sprintf "0x%x,0x90000000",$opcode{$1}; | |
448 | } | |
449 | last; | |
450 | }; | |
451 | /\.global|\.globl|\.extern/ | |
452 | && do { $globals{$line} = $prefix . $line; | |
453 | $line = $globals{$line} if ($prefix); | |
454 | last; | |
455 | }; | |
456 | /\.type/ && do { ($sym,$type,$narg) = split(',',$line); | |
457 | if ($type eq "\@function") { | |
458 | undef $current_function; | |
459 | $current_function->{name} = $sym; | |
460 | $current_function->{abi} = "svr4"; | |
461 | $current_function->{narg} = $narg; | |
462 | $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; | |
463 | } elsif ($type eq "\@abi-omnipotent") { | |
464 | undef $current_function; | |
465 | $current_function->{name} = $sym; | |
466 | $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE"; | |
467 | } | |
468 | $line =~ s/\@abi\-omnipotent/\@function/; | |
469 | $line =~ s/\@function.*/\@function/; | |
470 | last; | |
471 | }; | |
472 | /\.asciz/ && do { if ($line =~ /^"(.*)"$/) { | |
473 | $dir = ".byte"; | |
474 | $line = join(",",unpack("C*",$1),0); | |
475 | } | |
476 | last; | |
477 | }; | |
93c4ba07 AP |
478 | /\.rva|\.long|\.quad/ |
479 | && do { $line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei; | |
480 | $line =~ s/\.L/$decor/g; | |
481 | last; | |
482 | }; | |
85253772 AP |
483 | } |
484 | ||
485 | if ($gas) { | |
486 | $self->{value} = $dir . "\t" . $line; | |
487 | ||
488 | if ($dir =~ /\.extern/) { | |
932cc129 | 489 | $self->{value} = ""; # swallow extern |
85253772 AP |
490 | } elsif (!$elf && $dir =~ /\.type/) { |
491 | $self->{value} = ""; | |
492 | $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" . | |
493 | (defined($globals{$1})?".scl 2;":".scl 3;") . | |
494 | "\t.type 32;\t.endef" | |
495 | if ($win64 && $line =~ /([^,]+),\@function/); | |
496 | } elsif (!$elf && $dir =~ /\.size/) { | |
497 | $self->{value} = ""; | |
498 | if (defined($current_function)) { | |
93c4ba07 | 499 | $self->{value} .= "${decor}SEH_end_$current_function->{name}:" |
85253772 AP |
500 | if ($win64 && $current_function->{abi} eq "svr4"); |
501 | undef $current_function; | |
502 | } | |
503 | } elsif (!$elf && $dir =~ /\.align/) { | |
504 | $self->{value} = ".p2align\t" . (log($line)/log(2)); | |
505 | } elsif ($dir eq ".section") { | |
506 | $current_segment=$line; | |
507 | if (!$elf && $current_segment eq ".init") { | |
508 | if ($flavour eq "macosx") { $self->{value} = ".mod_init_func"; } | |
509 | elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.ctors"; } | |
510 | } | |
511 | } elsif ($dir =~ /\.(text|data)/) { | |
512 | $current_segment=".$1"; | |
ddc20d4d AP |
513 | } elsif ($dir =~ /\.hidden/) { |
514 | if ($flavour eq "macosx") { $self->{value} = ".private_extern\t$prefix$line"; } | |
515 | elsif ($flavour eq "mingw64") { $self->{value} = ""; } | |
516 | } elsif ($dir =~ /\.comm/) { | |
517 | $self->{value} = "$dir\t$prefix$line"; | |
ff6f9f96 | 518 | $self->{value} =~ s|,([0-9]+),([0-9]+)$|",$1,".log($2)/log(2)|e if ($flavour eq "macosx"); |
5d0d60e2 | 519 | } |
1cfd258e AP |
520 | $line = ""; |
521 | return $self; | |
522 | } | |
523 | ||
85253772 | 524 | # non-gas case or nasm/masm |
1cfd258e | 525 | SWITCH: for ($dir) { |
9b634c9b | 526 | /\.text/ && do { my $v=undef; |
a23e3dbe | 527 | if ($nasm) { |
9b634c9b | 528 | $v="section .text code align=64\n"; |
a23e3dbe AP |
529 | } else { |
530 | $v="$current_segment\tENDS\n" if ($current_segment); | |
9b634c9b | 531 | $current_segment = ".text\$"; |
a23e3dbe AP |
532 | $v.="$current_segment\tSEGMENT "; |
533 | $v.=$masm>=$masmref ? "ALIGN(64)" : "PAGE"; | |
534 | $v.=" 'CODE'"; | |
535 | } | |
1cfd258e AP |
536 | $self->{value} = $v; |
537 | last; | |
538 | }; | |
9b634c9b AP |
539 | /\.data/ && do { my $v=undef; |
540 | if ($nasm) { | |
541 | $v="section .data data align=8\n"; | |
542 | } else { | |
543 | $v="$current_segment\tENDS\n" if ($current_segment); | |
544 | $current_segment = "_DATA"; | |
545 | $v.="$current_segment\tSEGMENT"; | |
546 | } | |
547 | $self->{value} = $v; | |
548 | last; | |
549 | }; | |
550 | /\.section/ && do { my $v=undef; | |
85253772 AP |
551 | $line =~ s/([^,]*).*/$1/; |
552 | $line = ".CRT\$XCU" if ($line eq ".init"); | |
9b634c9b AP |
553 | if ($nasm) { |
554 | $v="section $line"; | |
555 | if ($line=~/\.([px])data/) { | |
556 | $v.=" rdata align="; | |
557 | $v.=$1 eq "p"? 4 : 8; | |
d6522548 AP |
558 | } elsif ($line=~/\.CRT\$/i) { |
559 | $v.=" rdata align=8"; | |
9b634c9b AP |
560 | } |
561 | } else { | |
562 | $v="$current_segment\tENDS\n" if ($current_segment); | |
563 | $v.="$line\tSEGMENT"; | |
564 | if ($line=~/\.([px])data/) { | |
565 | $v.=" READONLY"; | |
566 | $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref); | |
d6522548 | 567 | } elsif ($line=~/\.CRT\$/i) { |
e6903980 AP |
568 | $v.=" READONLY "; |
569 | $v.=$masm>=$masmref ? "ALIGN(8)" : "DWORD"; | |
9b634c9b AP |
570 | } |
571 | } | |
572 | $current_segment = $line; | |
573 | $self->{value} = $v; | |
574 | last; | |
575 | }; | |
a23e3dbe | 576 | /\.extern/ && do { $self->{value} = "EXTERN\t".$line; |
9b634c9b AP |
577 | $self->{value} .= ":NEAR" if ($masm); |
578 | last; | |
579 | }; | |
80aa9cc9 AP |
580 | /\.globl|.global/ |
581 | && do { $self->{value} = $masm?"PUBLIC":"global"; | |
582 | $self->{value} .= "\t".$line; | |
1cfd258e AP |
583 | last; |
584 | }; | |
5d0d60e2 | 585 | /\.size/ && do { if (defined($current_function)) { |
8fe8bae1 AP |
586 | undef $self->{value}; |
587 | if ($current_function->{abi} eq "svr4") { | |
93c4ba07 | 588 | $self->{value}="${decor}SEH_end_$current_function->{name}:"; |
8fe8bae1 AP |
589 | $self->{value}.=":\n" if($masm); |
590 | } | |
4010b341 | 591 | $self->{value}.="$current_function->{name}\tENDP" if($masm && $current_function->{name}); |
1cfd258e AP |
592 | undef $current_function; |
593 | } | |
594 | last; | |
595 | }; | |
596 | /\.align/ && do { $self->{value} = "ALIGN\t".$line; last; }; | |
85253772 | 597 | /\.(value|long|rva|quad)/ |
80aa9cc9 | 598 | && do { my $sz = substr($1,0,1); |
fead2539 | 599 | my @arr = split(/,\s*/,$line); |
1cfd258e | 600 | my $last = pop(@arr); |
a23e3dbe | 601 | my $conv = sub { my $var=shift; |
065c5d63 | 602 | $var=~s/^(0b[0-1]+)/oct($1)/eig; |
fead2539 | 603 | $var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm); |
85253772 | 604 | if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva")) |
80aa9cc9 | 605 | { $var=~s/([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; } |
9b634c9b | 606 | $var; |
a23e3dbe | 607 | }; |
1cfd258e | 608 | |
85253772 | 609 | $sz =~ tr/bvlrq/BWDDQ/; |
1cfd258e | 610 | $self->{value} = "\tD$sz\t"; |
a23e3dbe AP |
611 | for (@arr) { $self->{value} .= &$conv($_).","; } |
612 | $self->{value} .= &$conv($last); | |
1cfd258e AP |
613 | last; |
614 | }; | |
fead2539 | 615 | /\.byte/ && do { my @str=split(/,\s*/,$line); |
e8169520 | 616 | map(s/(0b[0-1]+)/oct($1)/eig,@str); |
bf785c98 | 617 | map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm); |
85253772 | 618 | while ($#str>15) { |
55eab3b7 | 619 | $self->{value}.="DB\t" |
85253772 AP |
620 | .join(",",@str[0..15])."\n"; |
621 | foreach (0..15) { shift @str; } | |
d68ff710 | 622 | } |
85253772 AP |
623 | $self->{value}.="DB\t" |
624 | .join(",",@str) if (@str); | |
d68ff710 AP |
625 | last; |
626 | }; | |
ddc20d4d AP |
627 | /\.comm/ && do { my @str=split(/,\s*/,$line); |
628 | my $v=undef; | |
629 | if ($nasm) { | |
301799b8 | 630 | $v.="common $prefix@str[0] @str[1]"; |
ddc20d4d AP |
631 | } else { |
632 | $v="$current_segment\tENDS\n" if ($current_segment); | |
94c64f9a | 633 | $current_segment = "_DATA"; |
ddc20d4d AP |
634 | $v.="$current_segment\tSEGMENT\n"; |
635 | $v.="COMM @str[0]:DWORD:".@str[1]/4; | |
636 | } | |
637 | $self->{value} = $v; | |
638 | last; | |
639 | }; | |
1cfd258e AP |
640 | } |
641 | $line = ""; | |
642 | } | |
643 | ||
644 | $ret; | |
645 | } | |
646 | sub out { | |
647 | my $self = shift; | |
648 | $self->{value}; | |
649 | } | |
650 | } | |
651 | ||
fead2539 AP |
652 | sub rex { |
653 | local *opcode=shift; | |
ddc20d4d AP |
654 | my ($dst,$src,$rex)=@_; |
655 | ||
656 | $rex|=0x04 if($dst>=8); | |
657 | $rex|=0x01 if($src>=8); | |
658 | push @opcode,($rex|0x40) if ($rex); | |
fead2539 AP |
659 | } |
660 | ||
a9e790b9 | 661 | # older gas and ml64 don't handle SSE>2 instructions |
fead2539 AP |
662 | my %regrm = ( "%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3, |
663 | "%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7 ); | |
664 | ||
c7b903e0 AP |
665 | my $movq = sub { # elderly gas can't handle inter-register movq |
666 | my $arg = shift; | |
667 | my @opcode=(0x66); | |
a87ff751 | 668 | if ($arg =~ /%xmm([0-9]+),\s*%r(\w+)/) { |
c7b903e0 AP |
669 | my ($src,$dst)=($1,$2); |
670 | if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } | |
671 | rex(\@opcode,$src,$dst,0x8); | |
672 | push @opcode,0x0f,0x7e; | |
673 | push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M | |
674 | @opcode; | |
a87ff751 | 675 | } elsif ($arg =~ /%r(\w+),\s*%xmm([0-9]+)/) { |
c7b903e0 AP |
676 | my ($src,$dst)=($2,$1); |
677 | if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } | |
678 | rex(\@opcode,$src,$dst,0x8); | |
679 | push @opcode,0x0f,0x6e; | |
680 | push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M | |
681 | @opcode; | |
682 | } else { | |
683 | (); | |
684 | } | |
685 | }; | |
686 | ||
fead2539 | 687 | my $pextrd = sub { |
a87ff751 | 688 | if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*(%\w+)/) { |
fead2539 AP |
689 | my @opcode=(0x66); |
690 | $imm=$1; | |
691 | $src=$2; | |
a9e790b9 | 692 | $dst=$3; |
fead2539 AP |
693 | if ($dst =~ /%r([0-9]+)d/) { $dst = $1; } |
694 | elsif ($dst =~ /%e/) { $dst = $regrm{$dst}; } | |
695 | rex(\@opcode,$src,$dst); | |
696 | push @opcode,0x0f,0x3a,0x16; | |
697 | push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M | |
698 | push @opcode,$imm; | |
a9e790b9 | 699 | @opcode; |
fead2539 | 700 | } else { |
a9e790b9 | 701 | (); |
fead2539 | 702 | } |
a9e790b9 | 703 | }; |
fead2539 AP |
704 | |
705 | my $pinsrd = sub { | |
a87ff751 | 706 | if (shift =~ /\$([0-9]+),\s*(%\w+),\s*%xmm([0-9]+)/) { |
fead2539 AP |
707 | my @opcode=(0x66); |
708 | $imm=$1; | |
a9e790b9 AP |
709 | $src=$2; |
710 | $dst=$3; | |
b5c6aab5 | 711 | if ($src =~ /%r([0-9]+)/) { $src = $1; } |
fead2539 AP |
712 | elsif ($src =~ /%e/) { $src = $regrm{$src}; } |
713 | rex(\@opcode,$dst,$src); | |
714 | push @opcode,0x0f,0x3a,0x22; | |
715 | push @opcode,0xc0|(($dst&7)<<3)|($src&7); # ModR/M | |
716 | push @opcode,$imm; | |
a9e790b9 | 717 | @opcode; |
fead2539 | 718 | } else { |
a9e790b9 | 719 | (); |
fead2539 | 720 | } |
a9e790b9 | 721 | }; |
fead2539 AP |
722 | |
723 | my $pshufb = sub { | |
a87ff751 | 724 | if (shift =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) { |
fead2539 | 725 | my @opcode=(0x66); |
a9e790b9 | 726 | rex(\@opcode,$2,$1); |
fead2539 | 727 | push @opcode,0x0f,0x38,0x00; |
a9e790b9 AP |
728 | push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M |
729 | @opcode; | |
fead2539 | 730 | } else { |
a9e790b9 | 731 | (); |
fead2539 | 732 | } |
a9e790b9 | 733 | }; |
fead2539 | 734 | |
b5c6aab5 | 735 | my $palignr = sub { |
a87ff751 | 736 | if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { |
b5c6aab5 AP |
737 | my @opcode=(0x66); |
738 | rex(\@opcode,$3,$2); | |
739 | push @opcode,0x0f,0x3a,0x0f; | |
740 | push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M | |
741 | push @opcode,$1; | |
742 | @opcode; | |
743 | } else { | |
744 | (); | |
745 | } | |
746 | }; | |
747 | ||
748 | my $pclmulqdq = sub { | |
749 | if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { | |
750 | my @opcode=(0x66); | |
751 | rex(\@opcode,$3,$2); | |
752 | push @opcode,0x0f,0x3a,0x44; | |
753 | push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M | |
754 | my $c=$1; | |
755 | push @opcode,$c=~/^0/?oct($c):$c; | |
756 | @opcode; | |
757 | } else { | |
758 | (); | |
759 | } | |
760 | }; | |
761 | ||
301799b8 AP |
762 | my $rdrand = sub { |
763 | if (shift =~ /%[er](\w+)/) { | |
764 | my @opcode=(); | |
765 | my $dst=$1; | |
766 | if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; } | |
767 | rex(\@opcode,0,$1,8); | |
768 | push @opcode,0x0f,0xc7,0xf0|($dst&7); | |
769 | @opcode; | |
770 | } else { | |
771 | (); | |
772 | } | |
773 | }; | |
774 | ||
f6ff1aa8 AP |
775 | sub rxb { |
776 | local *opcode=shift; | |
777 | my ($dst,$src1,$src2,$rxb)=@_; | |
778 | ||
779 | $rxb|=0x7<<5; | |
780 | $rxb&=~(0x04<<5) if($dst>=8); | |
781 | $rxb&=~(0x01<<5) if($src1>=8); | |
782 | $rxb&=~(0x02<<5) if($src2>=8); | |
783 | push @opcode,$rxb; | |
784 | } | |
785 | ||
786 | my $vprotd = sub { | |
787 | if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { | |
788 | my @opcode=(0x8f); | |
789 | rxb(\@opcode,$3,$2,-1,0x08); | |
790 | push @opcode,0x78,0xc2; | |
791 | push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M | |
792 | my $c=$1; | |
793 | push @opcode,$c=~/^0/?oct($c):$c; | |
794 | @opcode; | |
795 | } else { | |
796 | (); | |
797 | } | |
798 | }; | |
799 | ||
800 | my $vprotq = sub { | |
801 | if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { | |
802 | my @opcode=(0x8f); | |
803 | rxb(\@opcode,$3,$2,-1,0x08); | |
804 | push @opcode,0x78,0xc3; | |
805 | push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M | |
806 | my $c=$1; | |
807 | push @opcode,$c=~/^0/?oct($c):$c; | |
808 | @opcode; | |
809 | } else { | |
810 | (); | |
811 | } | |
812 | }; | |
813 | ||
9b634c9b AP |
814 | if ($nasm) { |
815 | print <<___; | |
816 | default rel | |
a9e790b9 | 817 | %define XMMWORD |
9b634c9b AP |
818 | ___ |
819 | } elsif ($masm) { | |
820 | print <<___; | |
821 | OPTION DOTNAME | |
822 | ___ | |
823 | } | |
1cfd258e AP |
824 | while($line=<>) { |
825 | ||
826 | chomp($line); | |
827 | ||
4b450519 AP |
828 | $line =~ s|[#!].*$||; # get rid of asm-style comments... |
829 | $line =~ s|/\*.*\*/||; # ... and C-style comments... | |
830 | $line =~ s|^\s+||; # ... and skip white spaces in beginning | |
1cfd258e AP |
831 | |
832 | undef $label; | |
833 | undef $opcode; | |
4db48824 | 834 | undef @args; |
1cfd258e AP |
835 | |
836 | if ($label=label->re(\$line)) { print $label->out(); } | |
837 | ||
838 | if (directive->re(\$line)) { | |
839 | printf "%s",directive->out(); | |
a9e790b9 AP |
840 | } elsif ($opcode=opcode->re(\$line)) { |
841 | my $asm = eval("\$".$opcode->mnemonic()); | |
842 | undef @bytes; | |
843 | ||
844 | if ((ref($asm) eq 'CODE') && scalar(@bytes=&$asm($line))) { | |
845 | print $gas?".byte\t":"DB\t",join(',',@bytes),"\n"; | |
846 | next; | |
847 | } | |
848 | ||
849 | ARGUMENT: while (1) { | |
4db48824 | 850 | my $arg; |
1cfd258e | 851 | |
4db48824 AP |
852 | if ($arg=register->re(\$line)) { opcode->size($arg->size()); } |
853 | elsif ($arg=const->re(\$line)) { } | |
854 | elsif ($arg=ea->re(\$line)) { } | |
855 | elsif ($arg=expr->re(\$line)) { } | |
856 | else { last ARGUMENT; } | |
1cfd258e | 857 | |
4db48824 | 858 | push @args,$arg; |
1cfd258e | 859 | |
4db48824 | 860 | last ARGUMENT if ($line !~ /^,/); |
1cfd258e | 861 | |
4db48824 | 862 | $line =~ s/^,\s*//; |
1cfd258e AP |
863 | } # ARGUMENT: |
864 | ||
4db48824 AP |
865 | if ($#args>=0) { |
866 | my $insn; | |
a9e790b9 AP |
867 | my $sz=opcode->size(); |
868 | ||
85253772 | 869 | if ($gas) { |
4db48824 | 870 | $insn = $opcode->out($#args>=1?$args[$#args]->size():$sz); |
fead2539 | 871 | @args = map($_->out($sz),@args); |
a9e790b9 | 872 | printf "\t%s\t%s",$insn,join(",",@args); |
5d0d60e2 | 873 | } else { |
4db48824 | 874 | $insn = $opcode->out(); |
046ea308 AP |
875 | foreach (@args) { |
876 | my $arg = $_->out(); | |
a9e790b9 AP |
877 | # $insn.=$sz compensates for movq, pinsrw, ... |
878 | if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; } | |
879 | if ($arg =~ /^mm[0-9]+$/) { $insn.=$sz; $sz="q" if(!$sz); last; } | |
046ea308 | 880 | } |
4db48824 | 881 | @args = reverse(@args); |
a23e3dbe | 882 | undef $sz if ($nasm && $opcode->mnemonic() eq "lea"); |
fead2539 | 883 | printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args)); |
1cfd258e | 884 | } |
1cfd258e AP |
885 | } else { |
886 | printf "\t%s",$opcode->out(); | |
887 | } | |
888 | } | |
889 | ||
890 | print $line,"\n"; | |
891 | } | |
892 | ||
85253772 AP |
893 | print "\n$current_segment\tENDS\n" if ($current_segment && $masm); |
894 | print "END\n" if ($masm); | |
1cfd258e AP |
895 | |
896 | close STDOUT; | |
897 | ||
80aa9cc9 | 898 | \f################################################# |
1cfd258e AP |
899 | # Cross-reference x86_64 ABI "card" |
900 | # | |
901 | # Unix Win64 | |
902 | # %rax * * | |
903 | # %rbx - - | |
904 | # %rcx #4 #1 | |
905 | # %rdx #3 #2 | |
906 | # %rsi #2 - | |
907 | # %rdi #1 - | |
908 | # %rbp - - | |
909 | # %rsp - - | |
910 | # %r8 #5 #3 | |
911 | # %r9 #6 #4 | |
912 | # %r10 * * | |
913 | # %r11 * * | |
914 | # %r12 - - | |
915 | # %r13 - - | |
916 | # %r14 - - | |
917 | # %r15 - - | |
918 | # | |
919 | # (*) volatile register | |
920 | # (-) preserved by callee | |
921 | # (#) Nth argument, volatile | |
922 | # | |
923 | # In Unix terms top of stack is argument transfer area for arguments | |
924 | # which could not be accomodated in registers. Or in other words 7th | |
925 | # [integer] argument resides at 8(%rsp) upon function entry point. | |
926 | # 128 bytes above %rsp constitute a "red zone" which is not touched | |
927 | # by signal handlers and can be used as temporal storage without | |
928 | # allocating a frame. | |
929 | # | |
930 | # In Win64 terms N*8 bytes on top of stack is argument transfer area, | |
931 | # which belongs to/can be overwritten by callee. N is the number of | |
932 | # arguments passed to callee, *but* not less than 4! This means that | |
933 | # upon function entry point 5th argument resides at 40(%rsp), as well | |
934 | # as that 32 bytes from 8(%rsp) can always be used as temporal | |
d256b957 AP |
935 | # storage [without allocating a frame]. One can actually argue that |
936 | # one can assume a "red zone" above stack pointer under Win64 as well. | |
6a3a7f30 AP |
937 | # Point is that at apparently no occasion Windows kernel would alter |
938 | # the area above user stack pointer in true asynchronous manner... | |
1cfd258e AP |
939 | # |
940 | # All the above means that if assembler programmer adheres to Unix | |
941 | # register and stack layout, but disregards the "red zone" existense, | |
942 | # it's possible to use following prologue and epilogue to "gear" from | |
943 | # Unix to Win64 ABI in leaf functions with not more than 6 arguments. | |
944 | # | |
945 | # omnipotent_function: | |
946 | # ifdef WIN64 | |
947 | # movq %rdi,8(%rsp) | |
948 | # movq %rsi,16(%rsp) | |
949 | # movq %rcx,%rdi ; if 1st argument is actually present | |
950 | # movq %rdx,%rsi ; if 2nd argument is actually ... | |
951 | # movq %r8,%rdx ; if 3rd argument is ... | |
952 | # movq %r9,%rcx ; if 4th argument ... | |
953 | # movq 40(%rsp),%r8 ; if 5th ... | |
954 | # movq 48(%rsp),%r9 ; if 6th ... | |
955 | # endif | |
956 | # ... | |
957 | # ifdef WIN64 | |
958 | # movq 8(%rsp),%rdi | |
959 | # movq 16(%rsp),%rsi | |
960 | # endif | |
961 | # ret | |
4f469342 | 962 | # |
80aa9cc9 AP |
963 | \f################################################# |
964 | # Win64 SEH, Structured Exception Handling. | |
965 | # | |
4f469342 AP |
966 | # Unlike on Unix systems(*) lack of Win64 stack unwinding information |
967 | # has undesired side-effect at run-time: if an exception is raised in | |
968 | # assembler subroutine such as those in question (basically we're | |
969 | # referring to segmentation violations caused by malformed input | |
970 | # parameters), the application is briskly terminated without invoking | |
971 | # any exception handlers, most notably without generating memory dump | |
972 | # or any user notification whatsoever. This poses a problem. It's | |
973 | # possible to address it by registering custom language-specific | |
974 | # handler that would restore processor context to the state at | |
975 | # subroutine entry point and return "exception is not handled, keep | |
976 | # unwinding" code. Writing such handler can be a challenge... But it's | |
977 | # doable, though requires certain coding convention. Consider following | |
978 | # snippet: | |
979 | # | |
80aa9cc9 | 980 | # .type function,@function |
4f469342 AP |
981 | # function: |
982 | # movq %rsp,%rax # copy rsp to volatile register | |
983 | # pushq %r15 # save non-volatile registers | |
984 | # pushq %rbx | |
985 | # pushq %rbp | |
986 | # movq %rsp,%r11 | |
987 | # subq %rdi,%r11 # prepare [variable] stack frame | |
988 | # andq $-64,%r11 | |
989 | # movq %rax,0(%r11) # check for exceptions | |
990 | # movq %r11,%rsp # allocate [variable] stack frame | |
991 | # movq %rax,0(%rsp) # save original rsp value | |
992 | # magic_point: | |
993 | # ... | |
994 | # movq 0(%rsp),%rcx # pull original rsp value | |
995 | # movq -24(%rcx),%rbp # restore non-volatile registers | |
996 | # movq -16(%rcx),%rbx | |
997 | # movq -8(%rcx),%r15 | |
998 | # movq %rcx,%rsp # restore original rsp | |
999 | # ret | |
80aa9cc9 | 1000 | # .size function,.-function |
4f469342 AP |
1001 | # |
1002 | # The key is that up to magic_point copy of original rsp value remains | |
1003 | # in chosen volatile register and no non-volatile register, except for | |
1004 | # rsp, is modified. While past magic_point rsp remains constant till | |
1005 | # the very end of the function. In this case custom language-specific | |
1006 | # exception handler would look like this: | |
1007 | # | |
1008 | # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, | |
1009 | # CONTEXT *context,DISPATCHER_CONTEXT *disp) | |
80aa9cc9 AP |
1010 | # { ULONG64 *rsp = (ULONG64 *)context->Rax; |
1011 | # if (context->Rip >= magic_point) | |
4f469342 AP |
1012 | # { rsp = ((ULONG64 **)context->Rsp)[0]; |
1013 | # context->Rbp = rsp[-3]; | |
1014 | # context->Rbx = rsp[-2]; | |
1015 | # context->R15 = rsp[-1]; | |
1016 | # } | |
1017 | # context->Rsp = (ULONG64)rsp; | |
1018 | # context->Rdi = rsp[1]; | |
1019 | # context->Rsi = rsp[2]; | |
1020 | # | |
1021 | # memcpy (disp->ContextRecord,context,sizeof(CONTEXT)); | |
1022 | # RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase, | |
1023 | # dips->ControlPc,disp->FunctionEntry,disp->ContextRecord, | |
1024 | # &disp->HandlerData,&disp->EstablisherFrame,NULL); | |
1025 | # return ExceptionContinueSearch; | |
1026 | # } | |
1027 | # | |
1028 | # It's appropriate to implement this handler in assembler, directly in | |
1029 | # function's module. In order to do that one has to know members' | |
1030 | # offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant | |
1031 | # values. Here they are: | |
1032 | # | |
1033 | # CONTEXT.Rax 120 | |
1034 | # CONTEXT.Rcx 128 | |
1035 | # CONTEXT.Rdx 136 | |
1036 | # CONTEXT.Rbx 144 | |
1037 | # CONTEXT.Rsp 152 | |
1038 | # CONTEXT.Rbp 160 | |
1039 | # CONTEXT.Rsi 168 | |
1040 | # CONTEXT.Rdi 176 | |
1041 | # CONTEXT.R8 184 | |
1042 | # CONTEXT.R9 192 | |
1043 | # CONTEXT.R10 200 | |
1044 | # CONTEXT.R11 208 | |
1045 | # CONTEXT.R12 216 | |
1046 | # CONTEXT.R13 224 | |
1047 | # CONTEXT.R14 232 | |
1048 | # CONTEXT.R15 240 | |
1049 | # CONTEXT.Rip 248 | |
6dd9066e | 1050 | # CONTEXT.Xmm6 512 |
4f469342 AP |
1051 | # sizeof(CONTEXT) 1232 |
1052 | # DISPATCHER_CONTEXT.ControlPc 0 | |
1053 | # DISPATCHER_CONTEXT.ImageBase 8 | |
1054 | # DISPATCHER_CONTEXT.FunctionEntry 16 | |
1055 | # DISPATCHER_CONTEXT.EstablisherFrame 24 | |
1056 | # DISPATCHER_CONTEXT.TargetIp 32 | |
1057 | # DISPATCHER_CONTEXT.ContextRecord 40 | |
1058 | # DISPATCHER_CONTEXT.LanguageHandler 48 | |
1059 | # DISPATCHER_CONTEXT.HandlerData 56 | |
1060 | # UNW_FLAG_NHANDLER 0 | |
1061 | # ExceptionContinueSearch 1 | |
1062 | # | |
80aa9cc9 AP |
1063 | # In order to tie the handler to the function one has to compose |
1064 | # couple of structures: one for .xdata segment and one for .pdata. | |
1065 | # | |
4f469342 | 1066 | # UNWIND_INFO structure for .xdata segment would be |
80aa9cc9 AP |
1067 | # |
1068 | # function_unwind_info: | |
1069 | # .byte 9,0,0,0 | |
85253772 | 1070 | # .rva handler |
80aa9cc9 AP |
1071 | # |
1072 | # This structure designates exception handler for a function with | |
1073 | # zero-length prologue, no stack frame or frame register. | |
1074 | # | |
1075 | # To facilitate composing of .pdata structures, auto-generated "gear" | |
1076 | # prologue copies rsp value to rax and denotes next instruction with | |
85253772 | 1077 | # .LSEH_begin_{function_name} label. This essentially defines the SEH |
80aa9cc9 AP |
1078 | # styling rule mentioned in the beginning. Position of this label is |
1079 | # chosen in such manner that possible exceptions raised in the "gear" | |
1080 | # prologue would be accounted to caller and unwound from latter's frame. | |
85253772 | 1081 | # End of function is marked with respective .LSEH_end_{function_name} |
80aa9cc9 AP |
1082 | # label. To summarize, .pdata segment would contain |
1083 | # | |
85253772 AP |
1084 | # .rva .LSEH_begin_function |
1085 | # .rva .LSEH_end_function | |
1086 | # .rva function_unwind_info | |
80aa9cc9 AP |
1087 | # |
1088 | # Reference to functon_unwind_info from .xdata segment is the anchor. | |
85253772 | 1089 | # In case you wonder why references are 32-bit .rvas and not 64-bit |
80aa9cc9 AP |
1090 | # .quads. References put into these two segments are required to be |
1091 | # *relative* to the base address of the current binary module, a.k.a. | |
1092 | # image base. No Win64 module, be it .exe or .dll, can be larger than | |
1093 | # 2GB and thus such relative references can be and are accommodated in | |
1094 | # 32 bits. | |
1095 | # | |
1096 | # Having reviewed the example function code, one can argue that "movq | |
1097 | # %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix | |
1098 | # rax would contain an undefined value. If this "offends" you, use | |
1099 | # another register and refrain from modifying rax till magic_point is | |
1100 | # reached, i.e. as if it was a non-volatile register. If more registers | |
1101 | # are required prior [variable] frame setup is completed, note that | |
1102 | # nobody says that you can have only one "magic point." You can | |
1103 | # "liberate" non-volatile registers by denoting last stack off-load | |
1104 | # instruction and reflecting it in finer grade unwind logic in handler. | |
1105 | # After all, isn't it why it's called *language-specific* handler... | |
1106 | # | |
1107 | # Attentive reader can notice that exceptions would be mishandled in | |
1108 | # auto-generated "gear" epilogue. Well, exception effectively can't | |
1109 | # occur there, because if memory area used by it was subject to | |
1110 | # segmentation violation, then it would be raised upon call to the | |
1111 | # function (and as already mentioned be accounted to caller, which is | |
1112 | # not a problem). If you're still not comfortable, then define tail | |
1113 | # "magic point" just prior ret instruction and have handler treat it... | |
4f469342 AP |
1114 | # |
1115 | # (*) Note that we're talking about run-time, not debug-time. Lack of | |
1116 | # unwind information makes debugging hard on both Windows and | |
1117 | # Unix. "Unlike" referes to the fact that on Unix signal handler | |
1118 | # will always be invoked, core dumped and appropriate exit code | |
1119 | # returned to parent (for user notification). |