summaryrefslogtreecommitdiff
path: root/src/lib/libcrypto/perlasm/x86_64-xlate.pl
diff options
context:
space:
mode:
authorcvs2svn <admin@example.com>2012-07-13 17:49:56 +0000
committercvs2svn <admin@example.com>2012-07-13 17:49:56 +0000
commitee04221ea8063435416c7e6369e6eae76843aa71 (patch)
tree821921a1dd0a5a3cece91121e121cc63c4b68128 /src/lib/libcrypto/perlasm/x86_64-xlate.pl
parentadf6731f6e1d04718aee00cb93435143046aee9a (diff)
downloadopenbsd-eric_g2k12.tar.gz
openbsd-eric_g2k12.tar.bz2
openbsd-eric_g2k12.zip
This commit was manufactured by cvs2git to create tag 'eric_g2k12'.eric_g2k12
Diffstat (limited to '')
-rwxr-xr-xsrc/lib/libcrypto/perlasm/x86_64-xlate.pl920
1 files changed, 0 insertions, 920 deletions
diff --git a/src/lib/libcrypto/perlasm/x86_64-xlate.pl b/src/lib/libcrypto/perlasm/x86_64-xlate.pl
deleted file mode 100755
index 68b4c1ca80..0000000000
--- a/src/lib/libcrypto/perlasm/x86_64-xlate.pl
+++ /dev/null
@@ -1,920 +0,0 @@
1#!/usr/bin/env perl
2
3# Ascetic x86_64 AT&T to MASM/NASM assembler translator by <appro>.
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;
23#
24# Dual-ABI styling rules.
25#
26# 1. Adhere to Unix register and stack layout [see cross-reference
27# ABI "card" at the end for explanation].
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
32# unified Win64 prologue and epilogue automatically. If you want
33# to take care of ABI differences yourself, tag functions as
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:-)
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.:
50#
51# .picmeup %rax
52# lea .Label-.(%rax),%rax
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.
57# 9. .init segment is allowed to contain calls to functions only.
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.
60
61my $flavour = shift;
62my $output = shift;
63if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
64
65{ my ($stddev,$stdino,@junk)=stat(STDOUT);
66 my ($outdev,$outino,@junk)=stat($output);
67
68 open STDOUT,">$output" || die "can't open $output: $!"
69 if (1 || $stddev!=$outdev || $stdino!=$outino);
70}
71
72my $gas=1; $gas=0 if ($output =~ /\.asm$/);
73my $elf=1; $elf=0 if (!$gas);
74my $win64=0;
75my $prefix="";
76my $decor=".L";
77
78my $masmref=8 + 50727*2**-32; # 8.00.50727 shipped with VS2005
79my $masm=0;
80my $PTR=" PTR";
81
82my $nasmref=2.03;
83my $nasm=0;
84
85if ($flavour eq "mingw64") { $gas=1; $elf=0; $win64=1;
86 $prefix=`echo __USER_LABEL_PREFIX__ | $ENV{CC} -E -P -`;
87 chomp($prefix);
88 }
89elsif ($flavour eq "macosx") { $gas=1; $elf=0; $prefix="_"; $decor="L\$"; }
90elsif ($flavour eq "masm") { $gas=0; $elf=0; $masm=$masmref; $win64=1; $decor="\$L\$"; }
91elsif ($flavour eq "nasm") { $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $decor="\$L\$"; $PTR=""; }
92elsif (!$gas)
93{ if ($ENV{ASM} =~ m/nasm/ && `nasm -v` =~ m/version ([0-9]+)\.([0-9]+)/i)
94 { $nasm = $1 + $2*0.01; $PTR=""; }
95 elsif (`ml64 2>&1` =~ m/Version ([0-9]+)\.([0-9]+)(\.([0-9]+))?/)
96 { $masm = $1 + $2*2**-16 + $4*2**-32; }
97 die "no assembler found on %PATH" if (!($nasm || $masm));
98 $win64=1;
99 $elf=0;
100 $decor="\$L\$";
101}
102
103my $current_segment;
104my $current_function;
105my %globals;
106
107{ package opcode; # pick up opcodes
108 sub re {
109 my $self = shift; # single instance in enough...
110 local *line = shift;
111 undef $ret;
112
113 if ($line =~ /^([a-z][a-z0-9]*)/i) {
114 $self->{op} = $1;
115 $ret = $self;
116 $line = substr($line,@+[0]); $line =~ s/^\s+//;
117
118 undef $self->{sz};
119 if ($self->{op} =~ /^(movz)b.*/) { # movz is pain...
120 $self->{op} = $1;
121 $self->{sz} = "b";
122 } elsif ($self->{op} =~ /call|jmp/) {
123 $self->{sz} = "";
124 } elsif ($self->{op} =~ /^p/ && $' !~ /^(ush|op)/) { # SSEn
125 $self->{sz} = "";
126 } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) {
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 $self = shift;
141 if ($gas) {
142 if ($self->{op} eq "movz") { # movz is pain...
143 sprintf "%s%s%s",$self->{op},$self->{sz},shift;
144 } elsif ($self->{op} =~ /^set/) {
145 "$self->{op}";
146 } elsif ($self->{op} eq "ret") {
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 && $current_segment eq ".init") {
154 ".p2align\t3\n\t.quad";
155 } else {
156 "$self->{op}$self->{sz}";
157 }
158 } else {
159 $self->{op} =~ s/^movz/movzx/;
160 if ($self->{op} eq "ret") {
161 $self->{op} = "";
162 if ($win64 && $current_function->{abi} eq "svr4") {
163 $self->{op} = "mov rdi,QWORD${PTR}[8+rsp]\t;WIN64 epilogue\n\t".
164 "mov rsi,QWORD${PTR}[16+rsp]\n\t";
165 }
166 $self->{op} .= "DB\t0F3h,0C3h\t\t;repret";
167 } elsif ($self->{op} =~ /^(pop|push)f/) {
168 $self->{op} .= $self->{sz};
169 } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") {
170 $self->{op} = "\tDQ";
171 }
172 $self->{op};
173 }
174 }
175 sub mnemonic {
176 my $self=shift;
177 my $op=shift;
178 $self->{op}=$op if (defined($op));
179 $self->{op};
180 }
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 $self = shift;
197
198 if ($gas) {
199 # Solaris /usr/ccs/bin/as can't handle multiplications
200 # in $self->{value}
201 $self->{value} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;
202 $self->{value} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg;
203 sprintf "\$%s",$self->{value};
204 } else {
205 $self->{value} =~ s/(0b[0-1]+)/oct($1)/eig;
206 $self->{value} =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm);
207 sprintf "%s",$self->{value};
208 }
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 $ret;
216
217 # optional * ---vvv--- appears in indirect jmp/call
218 if ($line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)/ &&
219 !($line =~ /^PIC_(GOT|PLT)/)) {
220 $self->{asterisk} = $1;
221 $self->{label} = $2;
222 ($self->{base},$self->{index},$self->{scale})=split(/,/,$3);
223 $self->{scale} = 1 if (!defined($self->{scale}));
224 $ret = $self;
225 $line = substr($line,@+[0]); $line =~ s/^\s+//;
226
227 if ($win64 && $self->{label} =~ s/\@GOTPCREL//) {
228 die if (opcode->mnemonic() ne "mov");
229 opcode->mnemonic("lea");
230 }
231 $self->{base} =~ s/^%//;
232 $self->{index} =~ s/^%// if (defined($self->{index}));
233 }
234 $ret;
235 }
236 sub size {}
237 sub out {
238 my $self = shift;
239 my $sz = shift;
240
241 $self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
242 $self->{label} =~ s/\.L/$decor/g;
243
244 # Silently convert all EAs to 64-bit. This is required for
245 # elder GNU assembler and results in more compact code,
246 # *but* most importantly AES module depends on this feature!
247 $self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
248 $self->{base} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
249
250 if ($gas) {
251 # Solaris /usr/ccs/bin/as can't handle multiplications
252 # in $self->{label}, new gas requires sign extension...
253 use integer;
254 $self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;
255 $self->{label} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg;
256 $self->{label} =~ s/([0-9]+)/$1<<32>>32/eg;
257 $self->{label} =~ s/^___imp_/__imp__/ if ($flavour eq "mingw64");
258
259 if (defined($self->{index})) {
260 sprintf "%s%s(%%%s,%%%s,%d)",$self->{asterisk},
261 $self->{label},$self->{base},
262 $self->{index},$self->{scale};
263 } else {
264 sprintf "%s%s(%%%s)", $self->{asterisk},$self->{label},$self->{base};
265 }
266 } else {
267 %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR", l=>"DWORD$PTR", q=>"QWORD$PTR" );
268
269 $self->{label} =~ s/\./\$/g;
270 $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig;
271 $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/);
272 $sz="q" if ($self->{asterisk});
273
274 if (defined($self->{index})) {
275 sprintf "%s[%s%s*%d+%s]",$szmap{$sz},
276 $self->{label}?"$self->{label}+":"",
277 $self->{index},$self->{scale},
278 $self->{base};
279 } elsif ($self->{base} eq "rip") {
280 sprintf "%s[%s]",$szmap{$sz},$self->{label};
281 } else {
282 sprintf "%s[%s%s]",$szmap{$sz},
283 $self->{label}?"$self->{label}+":"",
284 $self->{base};
285 }
286 }
287 }
288}
289{ package register; # pick up registers, which start with %.
290 sub re {
291 my $class = shift; # muliple instances...
292 my $self = {};
293 local *line = shift;
294 undef $ret;
295
296 # optional * ---vvv--- appears in indirect jmp/call
297 if ($line =~ /^(\*?)%(\w+)/) {
298 bless $self,$class;
299 $self->{asterisk} = $1;
300 $self->{value} = $2;
301 $ret = $self;
302 $line = substr($line,@+[0]); $line =~ s/^\s+//;
303 }
304 $ret;
305 }
306 sub size {
307 my $self = shift;
308 undef $ret;
309
310 if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; }
311 elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; }
312 elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; }
313 elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; }
314 elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; }
315 elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; }
316 elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; }
317 elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; }
318
319 $ret;
320 }
321 sub out {
322 my $self = shift;
323 if ($gas) { sprintf "%s%%%s",$self->{asterisk},$self->{value}; }
324 else { $self->{value}; }
325 }
326}
327{ package label; # pick up labels, which end with :
328 sub re {
329 my $self = shift; # single instance is enough...
330 local *line = shift;
331 undef $ret;
332
333 if ($line =~ /(^[\.\w]+)\:/) {
334 $self->{value} = $1;
335 $ret = $self;
336 $line = substr($line,@+[0]); $line =~ s/^\s+//;
337
338 $self->{value} =~ s/^\.L/$decor/;
339 }
340 $ret;
341 }
342 sub out {
343 my $self = shift;
344
345 if ($gas) {
346 my $func = ($globals{$self->{value}} or $self->{value}) . ":";
347 if ($win64 &&
348 $current_function->{name} eq $self->{value} &&
349 $current_function->{abi} eq "svr4") {
350 $func .= "\n";
351 $func .= " movq %rdi,8(%rsp)\n";
352 $func .= " movq %rsi,16(%rsp)\n";
353 $func .= " movq %rsp,%rax\n";
354 $func .= "${decor}SEH_begin_$current_function->{name}:\n";
355 my $narg = $current_function->{narg};
356 $narg=6 if (!defined($narg));
357 $func .= " movq %rcx,%rdi\n" if ($narg>0);
358 $func .= " movq %rdx,%rsi\n" if ($narg>1);
359 $func .= " movq %r8,%rdx\n" if ($narg>2);
360 $func .= " movq %r9,%rcx\n" if ($narg>3);
361 $func .= " movq 40(%rsp),%r8\n" if ($narg>4);
362 $func .= " movq 48(%rsp),%r9\n" if ($narg>5);
363 }
364 $func;
365 } elsif ($self->{value} ne "$current_function->{name}") {
366 $self->{value} .= ":" if ($masm && $ret!~m/^\$/);
367 $self->{value} . ":";
368 } elsif ($win64 && $current_function->{abi} eq "svr4") {
369 my $func = "$current_function->{name}" .
370 ($nasm ? ":" : "\tPROC $current_function->{scope}") .
371 "\n";
372 $func .= " mov QWORD${PTR}[8+rsp],rdi\t;WIN64 prologue\n";
373 $func .= " mov QWORD${PTR}[16+rsp],rsi\n";
374 $func .= " mov rax,rsp\n";
375 $func .= "${decor}SEH_begin_$current_function->{name}:";
376 $func .= ":" if ($masm);
377 $func .= "\n";
378 my $narg = $current_function->{narg};
379 $narg=6 if (!defined($narg));
380 $func .= " mov rdi,rcx\n" if ($narg>0);
381 $func .= " mov rsi,rdx\n" if ($narg>1);
382 $func .= " mov rdx,r8\n" if ($narg>2);
383 $func .= " mov rcx,r9\n" if ($narg>3);
384 $func .= " mov r8,QWORD${PTR}[40+rsp]\n" if ($narg>4);
385 $func .= " mov r9,QWORD${PTR}[48+rsp]\n" if ($narg>5);
386 $func .= "\n";
387 } else {
388 "$current_function->{name}".
389 ($nasm ? ":" : "\tPROC $current_function->{scope}");
390 }
391 }
392}
393{ package expr; # pick up expressioins
394 sub re {
395 my $self = shift; # single instance is enough...
396 local *line = shift;
397 undef $ret;
398
399 if ($line =~ /(^[^,]+)/) {
400 $self->{value} = $1;
401 $ret = $self;
402 $line = substr($line,@+[0]); $line =~ s/^\s+//;
403
404 $self->{value} =~ s/\@PLT// if (!$elf);
405 $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
406 $self->{value} =~ s/\.L/$decor/g;
407 }
408 $ret;
409 }
410 sub out {
411 my $self = shift;
412 if ($nasm && opcode->mnemonic()=~m/^j/) {
413 "NEAR ".$self->{value};
414 } else {
415 $self->{value};
416 }
417 }
418}
419{ package directive; # pick up directives, which start with .
420 sub re {
421 my $self = shift; # single instance is enough...
422 local *line = shift;
423 undef $ret;
424 my $dir;
425 my %opcode = # lea 2f-1f(%rip),%dst; 1: nop; 2:
426 ( "%rax"=>0x01058d48, "%rcx"=>0x010d8d48,
427 "%rdx"=>0x01158d48, "%rbx"=>0x011d8d48,
428 "%rsp"=>0x01258d48, "%rbp"=>0x012d8d48,
429 "%rsi"=>0x01358d48, "%rdi"=>0x013d8d48,
430 "%r8" =>0x01058d4c, "%r9" =>0x010d8d4c,
431 "%r10"=>0x01158d4c, "%r11"=>0x011d8d4c,
432 "%r12"=>0x01258d4c, "%r13"=>0x012d8d4c,
433 "%r14"=>0x01358d4c, "%r15"=>0x013d8d4c );
434
435 if ($line =~ /^\s*(\.\w+)/) {
436 $dir = $1;
437 $ret = $self;
438 undef $self->{value};
439 $line = substr($line,@+[0]); $line =~ s/^\s+//;
440
441 SWITCH: for ($dir) {
442 /\.picmeup/ && do { if ($line =~ /(%r[\w]+)/i) {
443 $dir="\t.long";
444 $line=sprintf "0x%x,0x90000000",$opcode{$1};
445 }
446 last;
447 };
448 /\.global|\.globl|\.extern/
449 && do { $globals{$line} = $prefix . $line;
450 $line = $globals{$line} if ($prefix);
451 last;
452 };
453 /\.type/ && do { ($sym,$type,$narg) = split(',',$line);
454 if ($type eq "\@function") {
455 undef $current_function;
456 $current_function->{name} = $sym;
457 $current_function->{abi} = "svr4";
458 $current_function->{narg} = $narg;
459 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";
460 } elsif ($type eq "\@abi-omnipotent") {
461 undef $current_function;
462 $current_function->{name} = $sym;
463 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";
464 }
465 $line =~ s/\@abi\-omnipotent/\@function/;
466 $line =~ s/\@function.*/\@function/;
467 last;
468 };
469 /\.asciz/ && do { if ($line =~ /^"(.*)"$/) {
470 $dir = ".byte";
471 $line = join(",",unpack("C*",$1),0);
472 }
473 last;
474 };
475 /\.rva|\.long|\.quad/
476 && do { $line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
477 $line =~ s/\.L/$decor/g;
478 last;
479 };
480 }
481
482 if ($gas) {
483 $self->{value} = $dir . "\t" . $line;
484
485 if ($dir =~ /\.extern/) {
486 $self->{value} = ""; # swallow extern
487 } elsif (!$elf && $dir =~ /\.type/) {
488 $self->{value} = "";
489 $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" .
490 (defined($globals{$1})?".scl 2;":".scl 3;") .
491 "\t.type 32;\t.endef"
492 if ($win64 && $line =~ /([^,]+),\@function/);
493 } elsif (!$elf && $dir =~ /\.size/) {
494 $self->{value} = "";
495 if (defined($current_function)) {
496 $self->{value} .= "${decor}SEH_end_$current_function->{name}:"
497 if ($win64 && $current_function->{abi} eq "svr4");
498 undef $current_function;
499 }
500 } elsif (!$elf && $dir =~ /\.align/) {
501 $self->{value} = ".p2align\t" . (log($line)/log(2));
502 } elsif ($dir eq ".section") {
503 $current_segment=$line;
504 if (!$elf && $current_segment eq ".init") {
505 if ($flavour eq "macosx") { $self->{value} = ".mod_init_func"; }
506 elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.ctors"; }
507 }
508 } elsif ($dir =~ /\.(text|data)/) {
509 $current_segment=".$1";
510 }
511 $line = "";
512 return $self;
513 }
514
515 # non-gas case or nasm/masm
516 SWITCH: for ($dir) {
517 /\.text/ && do { my $v=undef;
518 if ($nasm) {
519 $v="section .text code align=64\n";
520 } else {
521 $v="$current_segment\tENDS\n" if ($current_segment);
522 $current_segment = ".text\$";
523 $v.="$current_segment\tSEGMENT ";
524 $v.=$masm>=$masmref ? "ALIGN(64)" : "PAGE";
525 $v.=" 'CODE'";
526 }
527 $self->{value} = $v;
528 last;
529 };
530 /\.data/ && do { my $v=undef;
531 if ($nasm) {
532 $v="section .data data align=8\n";
533 } else {
534 $v="$current_segment\tENDS\n" if ($current_segment);
535 $current_segment = "_DATA";
536 $v.="$current_segment\tSEGMENT";
537 }
538 $self->{value} = $v;
539 last;
540 };
541 /\.section/ && do { my $v=undef;
542 $line =~ s/([^,]*).*/$1/;
543 $line = ".CRT\$XCU" if ($line eq ".init");
544 if ($nasm) {
545 $v="section $line";
546 if ($line=~/\.([px])data/) {
547 $v.=" rdata align=";
548 $v.=$1 eq "p"? 4 : 8;
549 } elsif ($line=~/\.CRT\$/i) {
550 $v.=" rdata align=8";
551 }
552 } else {
553 $v="$current_segment\tENDS\n" if ($current_segment);
554 $v.="$line\tSEGMENT";
555 if ($line=~/\.([px])data/) {
556 $v.=" READONLY";
557 $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref);
558 } elsif ($line=~/\.CRT\$/i) {
559 $v.=" READONLY DWORD";
560 }
561 }
562 $current_segment = $line;
563 $self->{value} = $v;
564 last;
565 };
566 /\.extern/ && do { $self->{value} = "EXTERN\t".$line;
567 $self->{value} .= ":NEAR" if ($masm);
568 last;
569 };
570 /\.globl|.global/
571 && do { $self->{value} = $masm?"PUBLIC":"global";
572 $self->{value} .= "\t".$line;
573 last;
574 };
575 /\.size/ && do { if (defined($current_function)) {
576 undef $self->{value};
577 if ($current_function->{abi} eq "svr4") {
578 $self->{value}="${decor}SEH_end_$current_function->{name}:";
579 $self->{value}.=":\n" if($masm);
580 }
581 $self->{value}.="$current_function->{name}\tENDP" if($masm);
582 undef $current_function;
583 }
584 last;
585 };
586 /\.align/ && do { $self->{value} = "ALIGN\t".$line; last; };
587 /\.(value|long|rva|quad)/
588 && do { my $sz = substr($1,0,1);
589 my @arr = split(/,\s*/,$line);
590 my $last = pop(@arr);
591 my $conv = sub { my $var=shift;
592 $var=~s/^(0b[0-1]+)/oct($1)/eig;
593 $var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm);
594 if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva"))
595 { $var=~s/([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; }
596 $var;
597 };
598
599 $sz =~ tr/bvlrq/BWDDQ/;
600 $self->{value} = "\tD$sz\t";
601 for (@arr) { $self->{value} .= &$conv($_).","; }
602 $self->{value} .= &$conv($last);
603 last;
604 };
605 /\.byte/ && do { my @str=split(/,\s*/,$line);
606 map(s/(0b[0-1]+)/oct($1)/eig,@str);
607 map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm);
608 while ($#str>15) {
609 $self->{value}.="DB\t"
610 .join(",",@str[0..15])."\n";
611 foreach (0..15) { shift @str; }
612 }
613 $self->{value}.="DB\t"
614 .join(",",@str) if (@str);
615 last;
616 };
617 }
618 $line = "";
619 }
620
621 $ret;
622 }
623 sub out {
624 my $self = shift;
625 $self->{value};
626 }
627}
628
629print "#include <machine/asm.h>\n";
630
631if ($nasm) {
632 print <<___;
633default rel
634___
635} elsif ($masm) {
636 print <<___;
637OPTION DOTNAME
638___
639}
640while($line=<>) {
641
642 chomp($line);
643
644 $line =~ s|[#!].*$||; # get rid of asm-style comments...
645 $line =~ s|/\*.*\*/||; # ... and C-style comments...
646 $line =~ s|^\s+||; # ... and skip white spaces in beginning
647
648 undef $label;
649 undef $opcode;
650 undef $sz;
651 undef @args;
652
653 if ($label=label->re(\$line)) { print $label->out(); }
654
655 if (directive->re(\$line)) {
656 printf "%s",directive->out();
657 } elsif ($opcode=opcode->re(\$line)) { ARGUMENT: while (1) {
658 my $arg;
659
660 if ($arg=register->re(\$line)) { opcode->size($arg->size()); }
661 elsif ($arg=const->re(\$line)) { }
662 elsif ($arg=ea->re(\$line)) { }
663 elsif ($arg=expr->re(\$line)) { }
664 else { last ARGUMENT; }
665
666 push @args,$arg;
667
668 last ARGUMENT if ($line !~ /^,/);
669
670 $line =~ s/^,\s*//;
671 } # ARGUMENT:
672
673 $sz=opcode->size();
674
675 if ($#args>=0) {
676 my $insn;
677 if ($gas) {
678 $insn = $opcode->out($#args>=1?$args[$#args]->size():$sz);
679 } else {
680 $insn = $opcode->out();
681 $insn .= $sz if (map($_->out() =~ /x?mm/,@args));
682 @args = reverse(@args);
683 undef $sz if ($nasm && $opcode->mnemonic() eq "lea");
684 }
685 printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args));
686 } else {
687 printf "\t%s",$opcode->out();
688 }
689 }
690
691 print $line,"\n";
692}
693
694print "\n$current_segment\tENDS\n" if ($current_segment && $masm);
695print "END\n" if ($masm);
696
697close STDOUT;
698
699 #################################################
700# Cross-reference x86_64 ABI "card"
701#
702# Unix Win64
703# %rax * *
704# %rbx - -
705# %rcx #4 #1
706# %rdx #3 #2
707# %rsi #2 -
708# %rdi #1 -
709# %rbp - -
710# %rsp - -
711# %r8 #5 #3
712# %r9 #6 #4
713# %r10 * *
714# %r11 * *
715# %r12 - -
716# %r13 - -
717# %r14 - -
718# %r15 - -
719#
720# (*) volatile register
721# (-) preserved by callee
722# (#) Nth argument, volatile
723#
724# In Unix terms top of stack is argument transfer area for arguments
725# which could not be accomodated in registers. Or in other words 7th
726# [integer] argument resides at 8(%rsp) upon function entry point.
727# 128 bytes above %rsp constitute a "red zone" which is not touched
728# by signal handlers and can be used as temporal storage without
729# allocating a frame.
730#
731# In Win64 terms N*8 bytes on top of stack is argument transfer area,
732# which belongs to/can be overwritten by callee. N is the number of
733# arguments passed to callee, *but* not less than 4! This means that
734# upon function entry point 5th argument resides at 40(%rsp), as well
735# as that 32 bytes from 8(%rsp) can always be used as temporal
736# storage [without allocating a frame]. One can actually argue that
737# one can assume a "red zone" above stack pointer under Win64 as well.
738# Point is that at apparently no occasion Windows kernel would alter
739# the area above user stack pointer in true asynchronous manner...
740#
741# All the above means that if assembler programmer adheres to Unix
742# register and stack layout, but disregards the "red zone" existense,
743# it's possible to use following prologue and epilogue to "gear" from
744# Unix to Win64 ABI in leaf functions with not more than 6 arguments.
745#
746# omnipotent_function:
747# ifdef WIN64
748# movq %rdi,8(%rsp)
749# movq %rsi,16(%rsp)
750# movq %rcx,%rdi ; if 1st argument is actually present
751# movq %rdx,%rsi ; if 2nd argument is actually ...
752# movq %r8,%rdx ; if 3rd argument is ...
753# movq %r9,%rcx ; if 4th argument ...
754# movq 40(%rsp),%r8 ; if 5th ...
755# movq 48(%rsp),%r9 ; if 6th ...
756# endif
757# ...
758# ifdef WIN64
759# movq 8(%rsp),%rdi
760# movq 16(%rsp),%rsi
761# endif
762# ret
763#
764 #################################################
765# Win64 SEH, Structured Exception Handling.
766#
767# Unlike on Unix systems(*) lack of Win64 stack unwinding information
768# has undesired side-effect at run-time: if an exception is raised in
769# assembler subroutine such as those in question (basically we're
770# referring to segmentation violations caused by malformed input
771# parameters), the application is briskly terminated without invoking
772# any exception handlers, most notably without generating memory dump
773# or any user notification whatsoever. This poses a problem. It's
774# possible to address it by registering custom language-specific
775# handler that would restore processor context to the state at
776# subroutine entry point and return "exception is not handled, keep
777# unwinding" code. Writing such handler can be a challenge... But it's
778# doable, though requires certain coding convention. Consider following
779# snippet:
780#
781# .type function,@function
782# function:
783# movq %rsp,%rax # copy rsp to volatile register
784# pushq %r15 # save non-volatile registers
785# pushq %rbx
786# pushq %rbp
787# movq %rsp,%r11
788# subq %rdi,%r11 # prepare [variable] stack frame
789# andq $-64,%r11
790# movq %rax,0(%r11) # check for exceptions
791# movq %r11,%rsp # allocate [variable] stack frame
792# movq %rax,0(%rsp) # save original rsp value
793# magic_point:
794# ...
795# movq 0(%rsp),%rcx # pull original rsp value
796# movq -24(%rcx),%rbp # restore non-volatile registers
797# movq -16(%rcx),%rbx
798# movq -8(%rcx),%r15
799# movq %rcx,%rsp # restore original rsp
800# ret
801# .size function,.-function
802#
803# The key is that up to magic_point copy of original rsp value remains
804# in chosen volatile register and no non-volatile register, except for
805# rsp, is modified. While past magic_point rsp remains constant till
806# the very end of the function. In this case custom language-specific
807# exception handler would look like this:
808#
809# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
810# CONTEXT *context,DISPATCHER_CONTEXT *disp)
811# { ULONG64 *rsp = (ULONG64 *)context->Rax;
812# if (context->Rip >= magic_point)
813# { rsp = ((ULONG64 **)context->Rsp)[0];
814# context->Rbp = rsp[-3];
815# context->Rbx = rsp[-2];
816# context->R15 = rsp[-1];
817# }
818# context->Rsp = (ULONG64)rsp;
819# context->Rdi = rsp[1];
820# context->Rsi = rsp[2];
821#
822# memcpy (disp->ContextRecord,context,sizeof(CONTEXT));
823# RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase,
824# dips->ControlPc,disp->FunctionEntry,disp->ContextRecord,
825# &disp->HandlerData,&disp->EstablisherFrame,NULL);
826# return ExceptionContinueSearch;
827# }
828#
829# It's appropriate to implement this handler in assembler, directly in
830# function's module. In order to do that one has to know members'
831# offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant
832# values. Here they are:
833#
834# CONTEXT.Rax 120
835# CONTEXT.Rcx 128
836# CONTEXT.Rdx 136
837# CONTEXT.Rbx 144
838# CONTEXT.Rsp 152
839# CONTEXT.Rbp 160
840# CONTEXT.Rsi 168
841# CONTEXT.Rdi 176
842# CONTEXT.R8 184
843# CONTEXT.R9 192
844# CONTEXT.R10 200
845# CONTEXT.R11 208
846# CONTEXT.R12 216
847# CONTEXT.R13 224
848# CONTEXT.R14 232
849# CONTEXT.R15 240
850# CONTEXT.Rip 248
851# CONTEXT.Xmm6 512
852# sizeof(CONTEXT) 1232
853# DISPATCHER_CONTEXT.ControlPc 0
854# DISPATCHER_CONTEXT.ImageBase 8
855# DISPATCHER_CONTEXT.FunctionEntry 16
856# DISPATCHER_CONTEXT.EstablisherFrame 24
857# DISPATCHER_CONTEXT.TargetIp 32
858# DISPATCHER_CONTEXT.ContextRecord 40
859# DISPATCHER_CONTEXT.LanguageHandler 48
860# DISPATCHER_CONTEXT.HandlerData 56
861# UNW_FLAG_NHANDLER 0
862# ExceptionContinueSearch 1
863#
864# In order to tie the handler to the function one has to compose
865# couple of structures: one for .xdata segment and one for .pdata.
866#
867# UNWIND_INFO structure for .xdata segment would be
868#
869# function_unwind_info:
870# .byte 9,0,0,0
871# .rva handler
872#
873# This structure designates exception handler for a function with
874# zero-length prologue, no stack frame or frame register.
875#
876# To facilitate composing of .pdata structures, auto-generated "gear"
877# prologue copies rsp value to rax and denotes next instruction with
878# .LSEH_begin_{function_name} label. This essentially defines the SEH
879# styling rule mentioned in the beginning. Position of this label is
880# chosen in such manner that possible exceptions raised in the "gear"
881# prologue would be accounted to caller and unwound from latter's frame.
882# End of function is marked with respective .LSEH_end_{function_name}
883# label. To summarize, .pdata segment would contain
884#
885# .rva .LSEH_begin_function
886# .rva .LSEH_end_function
887# .rva function_unwind_info
888#
889# Reference to functon_unwind_info from .xdata segment is the anchor.
890# In case you wonder why references are 32-bit .rvas and not 64-bit
891# .quads. References put into these two segments are required to be
892# *relative* to the base address of the current binary module, a.k.a.
893# image base. No Win64 module, be it .exe or .dll, can be larger than
894# 2GB and thus such relative references can be and are accommodated in
895# 32 bits.
896#
897# Having reviewed the example function code, one can argue that "movq
898# %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix
899# rax would contain an undefined value. If this "offends" you, use
900# another register and refrain from modifying rax till magic_point is
901# reached, i.e. as if it was a non-volatile register. If more registers
902# are required prior [variable] frame setup is completed, note that
903# nobody says that you can have only one "magic point." You can
904# "liberate" non-volatile registers by denoting last stack off-load
905# instruction and reflecting it in finer grade unwind logic in handler.
906# After all, isn't it why it's called *language-specific* handler...
907#
908# Attentive reader can notice that exceptions would be mishandled in
909# auto-generated "gear" epilogue. Well, exception effectively can't
910# occur there, because if memory area used by it was subject to
911# segmentation violation, then it would be raised upon call to the
912# function (and as already mentioned be accounted to caller, which is
913# not a problem). If you're still not comfortable, then define tail
914# "magic point" just prior ret instruction and have handler treat it...
915#
916# (*) Note that we're talking about run-time, not debug-time. Lack of
917# unwind information makes debugging hard on both Windows and
918# Unix. "Unlike" referes to the fact that on Unix signal handler
919# will always be invoked, core dumped and appropriate exit code
920# returned to parent (for user notification).
generated by cgit v1.2.3-55-g6feb (git 2.39.1) at 2025-11-24 05:28:47 +0000