]> git.ipfire.org Git - thirdparty/openssl.git/blob - crypto/perlasm/x86_64-xlate.pl
projects / thirdparty / openssl.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Update copyright year
[thirdparty/openssl.git] / crypto / perlasm / x86_64-xlate.pl
1 #! /usr/bin/env perl
2 # Copyright 2005-2020 The OpenSSL Project Authors. All Rights Reserved.
3 #
4 # Licensed under the Apache License 2.0 (the "License"). You may not use
5 # this file except in compliance with the License. You can obtain a copy
6 # in the file LICENSE in the source distribution or at
7 # https://www.openssl.org/source/license.html
8
9
10 # Ascetic x86_64 AT&T to MASM/NASM assembler translator by <appro>.
11 #
12 # Why AT&T to MASM and not vice versa? Several reasons. Because AT&T
13 # format is way easier to parse. Because it's simpler to "gear" from
14 # Unix ABI to Windows one [see cross-reference "card" at the end of
15 # file]. Because Linux targets were available first...
16 #
17 # In addition the script also "distills" code suitable for GNU
18 # assembler, so that it can be compiled with more rigid assemblers,
19 # such as Solaris /usr/ccs/bin/as.
20 #
21 # This translator is not designed to convert *arbitrary* assembler
22 # code from AT&T format to MASM one. It's designed to convert just
23 # enough to provide for dual-ABI OpenSSL modules development...
24 # There *are* limitations and you might have to modify your assembler
25 # code or this script to achieve the desired result...
26 #
27 # Currently recognized limitations:
28 #
29 # - can't use multiple ops per line;
30 #
31 # Dual-ABI styling rules.
32 #
33 # 1. Adhere to Unix register and stack layout [see cross-reference
34 # ABI "card" at the end for explanation].
35 # 2. Forget about "red zone," stick to more traditional blended
36 # stack frame allocation. If volatile storage is actually required
37 # that is. If not, just leave the stack as is.
38 # 3. Functions tagged with ".type name,@function" get crafted with
39 # unified Win64 prologue and epilogue automatically. If you want
40 # to take care of ABI differences yourself, tag functions as
41 # ".type name,@abi-omnipotent" instead.
42 # 4. To optimize the Win64 prologue you can specify number of input
43 # arguments as ".type name,@function,N." Keep in mind that if N is
44 # larger than 6, then you *have to* write "abi-omnipotent" code,
45 # because >6 cases can't be addressed with unified prologue.
46 # 5. Name local labels as .L*, do *not* use dynamic labels such as 1:
47 # (sorry about latter).
48 # 6. Don't use [or hand-code with .byte] "rep ret." "ret" mnemonic is
49 # required to identify the spots, where to inject Win64 epilogue!
50 # But on the pros, it's then prefixed with rep automatically:-)
51 # 7. Stick to explicit ip-relative addressing. If you have to use
52 # GOTPCREL addressing, stick to mov symbol@GOTPCREL(%rip),%r??.
53 # Both are recognized and translated to proper Win64 addressing
54 # modes.
55 #
56 # 8. In order to provide for structured exception handling unified
57 # Win64 prologue copies %rsp value to %rax. For further details
58 # see SEH paragraph at the end.
59 # 9. .init segment is allowed to contain calls to functions only.
60 # a. If function accepts more than 4 arguments *and* >4th argument
61 # is declared as non 64-bit value, do clear its upper part.
62 \f
63
64 use strict;
65
66 my $flavour = shift;
67 my $output = shift;
68 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
69
70 open STDOUT,">$output" || die "can't open $output: $!"
71 if (defined($output));
72
73 my $gas=1; $gas=0 if ($output =~ /\.asm$/);
74 my $elf=1; $elf=0 if (!$gas);
75 my $win64=0;
76 my $prefix="";
77 my $decor=".L";
78
79 my $masmref=8 + 50727*2**-32; # 8.00.50727 shipped with VS2005
80 my $masm=0;
81 my $PTR=" PTR";
82
83 my $nasmref=2.03;
84 my $nasm=0;
85
86 # GNU as indicator, as opposed to $gas, which indicates acceptable
87 # syntax
88 my $gnuas=0;
89
90 if ($flavour eq "mingw64") { $gas=1; $elf=0; $win64=1;
91 $prefix=`echo __USER_LABEL_PREFIX__ | $ENV{CC} -E -P -`;
92 $prefix =~ s|\R$||; # Better chomp
93 }
94 elsif ($flavour eq "macosx") { $gas=1; $elf=0; $prefix="_"; $decor="L\$"; }
95 elsif ($flavour eq "masm") { $gas=0; $elf=0; $masm=$masmref; $win64=1; $decor="\$L\$"; }
96 elsif ($flavour eq "nasm") { $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $decor="\$L\$"; $PTR=""; }
97 elsif (!$gas)
98 { if ($ENV{ASM} =~ m/nasm/ && `nasm -v` =~ m/version ([0-9]+)\.([0-9]+)/i)
99 { $nasm = $1 + $2*0.01; $PTR=""; }
100 elsif (`ml64 2>&1` =~ m/Version ([0-9]+)\.([0-9]+)(\.([0-9]+))?/)
101 { $masm = $1 + $2*2**-16 + $4*2**-32; }
102 die "no assembler found on %PATH%" if (!($nasm || $masm));
103 $win64=1;
104 $elf=0;
105 $decor="\$L\$";
106 }
107 # Find out if we're using GNU as
108 elsif (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
109 =~ /GNU assembler version ([2-9]\.[0-9]+)/)
110 {
111 $gnuas=1;
112 }
113 elsif (`$ENV{CC} --version 2>/dev/null`
114 =~ /clang .*/)
115 {
116 $gnuas=1;
117 }
118
119 my $cet_property;
120 if ($flavour =~ /elf/) {
121 # Always generate .note.gnu.property section for ELF outputs to
122 # mark Intel CET support since all input files must be marked
123 # with Intel CET support in order for linker to mark output with
124 # Intel CET support.
125 my $p2align=3; $p2align=2 if ($flavour eq "elf32");
126 my $section='.note.gnu.property, #alloc';
127 $section='".note.gnu.property", "a"' if $gnuas;
128 $cet_property = <<_____;
129 .section $section
130 .p2align $p2align
131 .long 1f - 0f
132 .long 4f - 1f
133 .long 5
134 0:
135 # "GNU" encoded with .byte, since .asciz isn't supported
136 # on Solaris.
137 .byte 0x47
138 .byte 0x4e
139 .byte 0x55
140 .byte 0
141 1:
142 .p2align $p2align
143 .long 0xc0000002
144 .long 3f - 2f
145 2:
146 .long 3
147 3:
148 .p2align $p2align
149 4:
150 _____
151 }
152
153 my $current_segment;
154 my $current_function;
155 my %globals;
156
157 { package opcode; # pick up opcodes
158 sub re {
159 my ($class, $line) = @_;
160 my $self = {};
161 my $ret;
162
163 if ($$line =~ /^([a-z][a-z0-9]*)/i) {
164 bless $self,$class;
165 $self->{op} = $1;
166 $ret = $self;
167 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
168
169 undef $self->{sz};
170 if ($self->{op} =~ /^(movz)x?([bw]).*/) { # movz is pain...
171 $self->{op} = $1;
172 $self->{sz} = $2;
173 } elsif ($self->{op} =~ /call|jmp/) {
174 $self->{sz} = "";
175 } elsif ($self->{op} =~ /^p/ && $' !~ /^(ush|op|insrw)/) { # SSEn
176 $self->{sz} = "";
177 } elsif ($self->{op} =~ /^[vk]/) { # VEX or k* such as kmov
178 $self->{sz} = "";
179 } elsif ($self->{op} =~ /mov[dq]/ && $$line =~ /%xmm/) {
180 $self->{sz} = "";
181 } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) {
182 $self->{op} = $1;
183 $self->{sz} = $2;
184 }
185 }
186 $ret;
187 }
188 sub size {
189 my ($self, $sz) = @_;
190 $self->{sz} = $sz if (defined($sz) && !defined($self->{sz}));
191 $self->{sz};
192 }
193 sub out {
194 my $self = shift;
195 if ($gas) {
196 if ($self->{op} eq "movz") { # movz is pain...
197 sprintf "%s%s%s",$self->{op},$self->{sz},shift;
198 } elsif ($self->{op} =~ /^set/) {
199 "$self->{op}";
200 } elsif ($self->{op} eq "ret") {
201 my $epilogue = "";
202 if ($win64 && $current_function->{abi} eq "svr4") {
203 $epilogue = "movq 8(%rsp),%rdi\n\t" .
204 "movq 16(%rsp),%rsi\n\t";
205 }
206 $epilogue . ".byte 0xf3,0xc3";
207 } elsif ($self->{op} eq "call" && !$elf && $current_segment eq ".init") {
208 ".p2align\t3\n\t.quad";
209 } else {
210 "$self->{op}$self->{sz}";
211 }
212 } else {
213 $self->{op} =~ s/^movz/movzx/;
214 if ($self->{op} eq "ret") {
215 $self->{op} = "";
216 if ($win64 && $current_function->{abi} eq "svr4") {
217 $self->{op} = "mov rdi,QWORD$PTR\[8+rsp\]\t;WIN64 epilogue\n\t".
218 "mov rsi,QWORD$PTR\[16+rsp\]\n\t";
219 }
220 $self->{op} .= "DB\t0F3h,0C3h\t\t;repret";
221 } elsif ($self->{op} =~ /^(pop|push)f/) {
222 $self->{op} .= $self->{sz};
223 } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") {
224 $self->{op} = "\tDQ";
225 }
226 $self->{op};
227 }
228 }
229 sub mnemonic {
230 my ($self, $op) = @_;
231 $self->{op}=$op if (defined($op));
232 $self->{op};
233 }
234 }
235 { package const; # pick up constants, which start with $
236 sub re {
237 my ($class, $line) = @_;
238 my $self = {};
239 my $ret;
240
241 if ($$line =~ /^\$([^,]+)/) {
242 bless $self, $class;
243 $self->{value} = $1;
244 $ret = $self;
245 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
246 }
247 $ret;
248 }
249 sub out {
250 my $self = shift;
251
252 $self->{value} =~ s/\b(0b[0-1]+)/oct($1)/eig;
253 if ($gas) {
254 # Solaris /usr/ccs/bin/as can't handle multiplications
255 # in $self->{value}
256 my $value = $self->{value};
257 no warnings; # oct might complain about overflow, ignore here...
258 $value =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;
259 if ($value =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg) {
260 $self->{value} = $value;
261 }
262 sprintf "\$%s",$self->{value};
263 } else {
264 my $value = $self->{value};
265 $value =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm);
266 sprintf "%s",$value;
267 }
268 }
269 }
270 { package ea; # pick up effective addresses: expr(%reg,%reg,scale)
271
272 my %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR",
273 l=>"DWORD$PTR", d=>"DWORD$PTR",
274 q=>"QWORD$PTR", o=>"OWORD$PTR",
275 x=>"XMMWORD$PTR", y=>"YMMWORD$PTR",
276 z=>"ZMMWORD$PTR" ) if (!$gas);
277
278 sub re {
279 my ($class, $line, $opcode) = @_;
280 my $self = {};
281 my $ret;
282
283 # optional * ----vvv--- appears in indirect jmp/call
284 if ($$line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)((?:{[^}]+})*)/) {
285 bless $self, $class;
286 $self->{asterisk} = $1;
287 $self->{label} = $2;
288 ($self->{base},$self->{index},$self->{scale})=split(/,/,$3);
289 $self->{scale} = 1 if (!defined($self->{scale}));
290 $self->{opmask} = $4;
291 $ret = $self;
292 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
293
294 if ($win64 && $self->{label} =~ s/\@GOTPCREL//) {
295 die if ($opcode->mnemonic() ne "mov");
296 $opcode->mnemonic("lea");
297 }
298 $self->{base} =~ s/^%//;
299 $self->{index} =~ s/^%// if (defined($self->{index}));
300 $self->{opcode} = $opcode;
301 }
302 $ret;
303 }
304 sub size {}
305 sub out {
306 my ($self, $sz) = @_;
307
308 $self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
309 $self->{label} =~ s/\.L/$decor/g;
310
311 # Silently convert all EAs to 64-bit. This is required for
312 # elder GNU assembler and results in more compact code,
313 # *but* most importantly AES module depends on this feature!
314 $self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
315 $self->{base} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
316
317 # Solaris /usr/ccs/bin/as can't handle multiplications
318 # in $self->{label}...
319 use integer;
320 $self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;
321 $self->{label} =~ s/\b([0-9]+\s*[\*\/\%]\s*[0-9]+)\b/eval($1)/eg;
322
323 # Some assemblers insist on signed presentation of 32-bit
324 # offsets, but sign extension is a tricky business in perl...
325 if ((1<<31)<<1) {
326 $self->{label} =~ s/\b([0-9]+)\b/$1<<32>>32/eg;
327 } else {
328 $self->{label} =~ s/\b([0-9]+)\b/$1>>0/eg;
329 }
330
331 # if base register is %rbp or %r13, see if it's possible to
332 # flip base and index registers [for better performance]
333 if (!$self->{label} && $self->{index} && $self->{scale}==1 &&
334 $self->{base} =~ /(rbp|r13)/) {
335 $self->{base} = $self->{index}; $self->{index} = $1;
336 }
337
338 if ($gas) {
339 $self->{label} =~ s/^___imp_/__imp__/ if ($flavour eq "mingw64");
340
341 if (defined($self->{index})) {
342 sprintf "%s%s(%s,%%%s,%d)%s",
343 $self->{asterisk},$self->{label},
344 $self->{base}?"%$self->{base}":"",
345 $self->{index},$self->{scale},
346 $self->{opmask};
347 } else {
348 sprintf "%s%s(%%%s)%s", $self->{asterisk},$self->{label},
349 $self->{base},$self->{opmask};
350 }
351 } else {
352 $self->{label} =~ s/\./\$/g;
353 $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig;
354 $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/);
355
356 my $mnemonic = $self->{opcode}->mnemonic();
357 ($self->{asterisk}) && ($sz="q") ||
358 ($mnemonic =~ /^v?mov([qd])$/) && ($sz=$1) ||
359 ($mnemonic =~ /^v?pinsr([qdwb])$/) && ($sz=$1) ||
360 ($mnemonic =~ /^vpbroadcast([qdwb])$/) && ($sz=$1) ||
361 ($mnemonic =~ /^v(?!perm)[a-z]+[fi]128$/) && ($sz="x");
362
363 $self->{opmask} =~ s/%(k[0-7])/$1/;
364
365 if (defined($self->{index})) {
366 sprintf "%s[%s%s*%d%s]%s",$szmap{$sz},
367 $self->{label}?"$self->{label}+":"",
368 $self->{index},$self->{scale},
369 $self->{base}?"+$self->{base}":"",
370 $self->{opmask};
371 } elsif ($self->{base} eq "rip") {
372 sprintf "%s[%s]",$szmap{$sz},$self->{label};
373 } else {
374 sprintf "%s[%s%s]%s", $szmap{$sz},
375 $self->{label}?"$self->{label}+":"",
376 $self->{base},$self->{opmask};
377 }
378 }
379 }
380 }
381 { package register; # pick up registers, which start with %.
382 sub re {
383 my ($class, $line, $opcode) = @_;
384 my $self = {};
385 my $ret;
386
387 # optional * ----vvv--- appears in indirect jmp/call
388 if ($$line =~ /^(\*?)%(\w+)((?:{[^}]+})*)/) {
389 bless $self,$class;
390 $self->{asterisk} = $1;
391 $self->{value} = $2;
392 $self->{opmask} = $3;
393 $opcode->size($self->size());
394 $ret = $self;
395 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
396 }
397 $ret;
398 }
399 sub size {
400 my $self = shift;
401 my $ret;
402
403 if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; }
404 elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; }
405 elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; }
406 elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; }
407 elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; }
408 elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; }
409 elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; }
410 elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; }
411
412 $ret;
413 }
414 sub out {
415 my $self = shift;
416 if ($gas) { sprintf "%s%%%s%s", $self->{asterisk},
417 $self->{value},
418 $self->{opmask}; }
419 else { $self->{opmask} =~ s/%(k[0-7])/$1/;
420 $self->{value}.$self->{opmask}; }
421 }
422 }
423 { package label; # pick up labels, which end with :
424 sub re {
425 my ($class, $line) = @_;
426 my $self = {};
427 my $ret;
428
429 if ($$line =~ /(^[\.\w]+)\:/) {
430 bless $self,$class;
431 $self->{value} = $1;
432 $ret = $self;
433 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
434
435 $self->{value} =~ s/^\.L/$decor/;
436 }
437 $ret;
438 }
439 sub out {
440 my $self = shift;
441
442 if ($gas) {
443 my $func = ($globals{$self->{value}} or $self->{value}) . ":";
444 if ($win64 && $current_function->{name} eq $self->{value}
445 && $current_function->{abi} eq "svr4") {
446 $func .= "\n";
447 $func .= " movq %rdi,8(%rsp)\n";
448 $func .= " movq %rsi,16(%rsp)\n";
449 $func .= " movq %rsp,%rax\n";
450 $func .= "${decor}SEH_begin_$current_function->{name}:\n";
451 my $narg = $current_function->{narg};
452 $narg=6 if (!defined($narg));
453 $func .= " movq %rcx,%rdi\n" if ($narg>0);
454 $func .= " movq %rdx,%rsi\n" if ($narg>1);
455 $func .= " movq %r8,%rdx\n" if ($narg>2);
456 $func .= " movq %r9,%rcx\n" if ($narg>3);
457 $func .= " movq 40(%rsp),%r8\n" if ($narg>4);
458 $func .= " movq 48(%rsp),%r9\n" if ($narg>5);
459 }
460 $func;
461 } elsif ($self->{value} ne "$current_function->{name}") {
462 # Make all labels in masm global.
463 $self->{value} .= ":" if ($masm);
464 $self->{value} . ":";
465 } elsif ($win64 && $current_function->{abi} eq "svr4") {
466 my $func = "$current_function->{name}" .
467 ($nasm ? ":" : "\tPROC $current_function->{scope}") .
468 "\n";
469 $func .= " mov QWORD$PTR\[8+rsp\],rdi\t;WIN64 prologue\n";
470 $func .= " mov QWORD$PTR\[16+rsp\],rsi\n";
471 $func .= " mov rax,rsp\n";
472 $func .= "${decor}SEH_begin_$current_function->{name}:";
473 $func .= ":" if ($masm);
474 $func .= "\n";
475 my $narg = $current_function->{narg};
476 $narg=6 if (!defined($narg));
477 $func .= " mov rdi,rcx\n" if ($narg>0);
478 $func .= " mov rsi,rdx\n" if ($narg>1);
479 $func .= " mov rdx,r8\n" if ($narg>2);
480 $func .= " mov rcx,r9\n" if ($narg>3);
481 $func .= " mov r8,QWORD$PTR\[40+rsp\]\n" if ($narg>4);
482 $func .= " mov r9,QWORD$PTR\[48+rsp\]\n" if ($narg>5);
483 $func .= "\n";
484 } else {
485 "$current_function->{name}".
486 ($nasm ? ":" : "\tPROC $current_function->{scope}");
487 }
488 }
489 }
490 { package expr; # pick up expressions
491 sub re {
492 my ($class, $line, $opcode) = @_;
493 my $self = {};
494 my $ret;
495
496 if ($$line =~ /(^[^,]+)/) {
497 bless $self,$class;
498 $self->{value} = $1;
499 $ret = $self;
500 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
501
502 $self->{value} =~ s/\@PLT// if (!$elf);
503 $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
504 $self->{value} =~ s/\.L/$decor/g;
505 $self->{opcode} = $opcode;
506 }
507 $ret;
508 }
509 sub out {
510 my $self = shift;
511 if ($nasm && $self->{opcode}->mnemonic()=~m/^j(?![re]cxz)/) {
512 "NEAR ".$self->{value};
513 } else {
514 $self->{value};
515 }
516 }
517 }
518 { package cfi_directive;
519 # CFI directives annotate instructions that are significant for
520 # stack unwinding procedure compliant with DWARF specification,
521 # see http://dwarfstd.org/. Besides naturally expected for this
522 # script platform-specific filtering function, this module adds
523 # three auxiliary synthetic directives not recognized by [GNU]
524 # assembler:
525 #
526 # - .cfi_push to annotate push instructions in prologue, which
527 # translates to .cfi_adjust_cfa_offset (if needed) and
528 # .cfi_offset;
529 # - .cfi_pop to annotate pop instructions in epilogue, which
530 # translates to .cfi_adjust_cfa_offset (if needed) and
531 # .cfi_restore;
532 # - [and most notably] .cfi_cfa_expression which encodes
533 # DW_CFA_def_cfa_expression and passes it to .cfi_escape as
534 # byte vector;
535 #
536 # CFA expressions were introduced in DWARF specification version
537 # 3 and describe how to deduce CFA, Canonical Frame Address. This
538 # becomes handy if your stack frame is variable and you can't
539 # spare register for [previous] frame pointer. Suggested directive
540 # syntax is made-up mix of DWARF operator suffixes [subset of]
541 # and references to registers with optional bias. Following example
542 # describes offloaded *original* stack pointer at specific offset
543 # from *current* stack pointer:
544 #
545 # .cfi_cfa_expression %rsp+40,deref,+8
546 #
547 # Final +8 has everything to do with the fact that CFA is defined
548 # as reference to top of caller's stack, and on x86_64 call to
549 # subroutine pushes 8-byte return address. In other words original
550 # stack pointer upon entry to a subroutine is 8 bytes off from CFA.
551
552 # Below constants are taken from "DWARF Expressions" section of the
553 # DWARF specification, section is numbered 7.7 in versions 3 and 4.
554 my %DW_OP_simple = ( # no-arg operators, mapped directly
555 deref => 0x06, dup => 0x12,
556 drop => 0x13, over => 0x14,
557 pick => 0x15, swap => 0x16,
558 rot => 0x17, xderef => 0x18,
559
560 abs => 0x19, and => 0x1a,
561 div => 0x1b, minus => 0x1c,
562 mod => 0x1d, mul => 0x1e,
563 neg => 0x1f, not => 0x20,
564 or => 0x21, plus => 0x22,
565 shl => 0x24, shr => 0x25,
566 shra => 0x26, xor => 0x27,
567 );
568
569 my %DW_OP_complex = ( # used in specific subroutines
570 constu => 0x10, # uleb128
571 consts => 0x11, # sleb128
572 plus_uconst => 0x23, # uleb128
573 lit0 => 0x30, # add 0-31 to opcode
574 reg0 => 0x50, # add 0-31 to opcode
575 breg0 => 0x70, # add 0-31 to opcole, sleb128
576 regx => 0x90, # uleb28
577 fbreg => 0x91, # sleb128
578 bregx => 0x92, # uleb128, sleb128
579 piece => 0x93, # uleb128
580 );
581
582 # Following constants are defined in x86_64 ABI supplement, for
583 # example available at https://www.uclibc.org/docs/psABI-x86_64.pdf,
584 # see section 3.7 "Stack Unwind Algorithm".
585 my %DW_reg_idx = (
586 "%rax"=>0, "%rdx"=>1, "%rcx"=>2, "%rbx"=>3,
587 "%rsi"=>4, "%rdi"=>5, "%rbp"=>6, "%rsp"=>7,
588 "%r8" =>8, "%r9" =>9, "%r10"=>10, "%r11"=>11,
589 "%r12"=>12, "%r13"=>13, "%r14"=>14, "%r15"=>15
590 );
591
592 my ($cfa_reg, $cfa_rsp);
593 my @cfa_stack;
594
595 # [us]leb128 format is variable-length integer representation base
596 # 2^128, with most significant bit of each byte being 0 denoting
597 # *last* most significant digit. See "Variable Length Data" in the
598 # DWARF specification, numbered 7.6 at least in versions 3 and 4.
599 sub sleb128 {
600 use integer; # get right shift extend sign
601
602 my $val = shift;
603 my $sign = ($val < 0) ? -1 : 0;
604 my @ret = ();
605
606 while(1) {
607 push @ret, $val&0x7f;
608
609 # see if remaining bits are same and equal to most
610 # significant bit of the current digit, if so, it's
611 # last digit...
612 last if (($val>>6) == $sign);
613
614 @ret[-1] |= 0x80;
615 $val >>= 7;
616 }
617
618 return @ret;
619 }
620 sub uleb128 {
621 my $val = shift;
622 my @ret = ();
623
624 while(1) {
625 push @ret, $val&0x7f;
626
627 # see if it's last significant digit...
628 last if (($val >>= 7) == 0);
629
630 @ret[-1] |= 0x80;
631 }
632
633 return @ret;
634 }
635 sub const {
636 my $val = shift;
637
638 if ($val >= 0 && $val < 32) {
639 return ($DW_OP_complex{lit0}+$val);
640 }
641 return ($DW_OP_complex{consts}, sleb128($val));
642 }
643 sub reg {
644 my $val = shift;
645
646 return if ($val !~ m/^(%r\w+)(?:([\+\-])((?:0x)?[0-9a-f]+))?/);
647
648 my $reg = $DW_reg_idx{$1};
649 my $off = eval ("0 $2 $3");
650
651 return (($DW_OP_complex{breg0} + $reg), sleb128($off));
652 # Yes, we use DW_OP_bregX+0 to push register value and not
653 # DW_OP_regX, because latter would require even DW_OP_piece,
654 # which would be a waste under the circumstances. If you have
655 # to use DWP_OP_reg, use "regx:N"...
656 }
657 sub cfa_expression {
658 my $line = shift;
659 my @ret;
660
661 foreach my $token (split(/,\s*/,$line)) {
662 if ($token =~ /^%r/) {
663 push @ret,reg($token);
664 } elsif ($token =~ /((?:0x)?[0-9a-f]+)\((%r\w+)\)/) {
665 push @ret,reg("$2+$1");
666 } elsif ($token =~ /(\w+):(\-?(?:0x)?[0-9a-f]+)(U?)/i) {
667 my $i = 1*eval($2);
668 push @ret,$DW_OP_complex{$1}, ($3 ? uleb128($i) : sleb128($i));
669 } elsif (my $i = 1*eval($token) or $token eq "0") {
670 if ($token =~ /^\+/) {
671 push @ret,$DW_OP_complex{plus_uconst},uleb128($i);
672 } else {
673 push @ret,const($i);
674 }
675 } else {
676 push @ret,$DW_OP_simple{$token};
677 }
678 }
679
680 # Finally we return DW_CFA_def_cfa_expression, 15, followed by
681 # length of the expression and of course the expression itself.
682 return (15,scalar(@ret),@ret);
683 }
684 sub re {
685 my ($class, $line) = @_;
686 my $self = {};
687 my $ret;
688
689 if ($$line =~ s/^\s*\.cfi_(\w+)\s*//) {
690 bless $self,$class;
691 $ret = $self;
692 undef $self->{value};
693 my $dir = $1;
694
695 SWITCH: for ($dir) {
696 # What is $cfa_rsp? Effectively it's difference between %rsp
697 # value and current CFA, Canonical Frame Address, which is
698 # why it starts with -8. Recall that CFA is top of caller's
699 # stack...
700 /startproc/ && do { ($cfa_reg, $cfa_rsp) = ("%rsp", -8); last; };
701 /endproc/ && do { ($cfa_reg, $cfa_rsp) = ("%rsp", 0);
702 # .cfi_remember_state directives that are not
703 # matched with .cfi_restore_state are
704 # unnecessary.
705 die "unpaired .cfi_remember_state" if (@cfa_stack);
706 last;
707 };
708 /def_cfa_register/
709 && do { $cfa_reg = $$line; last; };
710 /def_cfa_offset/
711 && do { $cfa_rsp = -1*eval($$line) if ($cfa_reg eq "%rsp");
712 last;
713 };
714 /adjust_cfa_offset/
715 && do { $cfa_rsp -= 1*eval($$line) if ($cfa_reg eq "%rsp");
716 last;
717 };
718 /def_cfa/ && do { if ($$line =~ /(%r\w+)\s*,\s*(.+)/) {
719 $cfa_reg = $1;
720 $cfa_rsp = -1*eval($2) if ($cfa_reg eq "%rsp");
721 }
722 last;
723 };
724 /push/ && do { $dir = undef;
725 $cfa_rsp -= 8;
726 if ($cfa_reg eq "%rsp") {
727 $self->{value} = ".cfi_adjust_cfa_offset\t8\n";
728 }
729 $self->{value} .= ".cfi_offset\t$$line,$cfa_rsp";
730 last;
731 };
732 /pop/ && do { $dir = undef;
733 $cfa_rsp += 8;
734 if ($cfa_reg eq "%rsp") {
735 $self->{value} = ".cfi_adjust_cfa_offset\t-8\n";
736 }
737 $self->{value} .= ".cfi_restore\t$$line";
738 last;
739 };
740 /cfa_expression/
741 && do { $dir = undef;
742 $self->{value} = ".cfi_escape\t" .
743 join(",", map(sprintf("0x%02x", $_),
744 cfa_expression($$line)));
745 last;
746 };
747 /remember_state/
748 && do { push @cfa_stack, [$cfa_reg, $cfa_rsp];
749 last;
750 };
751 /restore_state/
752 && do { ($cfa_reg, $cfa_rsp) = @{pop @cfa_stack};
753 last;
754 };
755 }
756
757 $self->{value} = ".cfi_$dir\t$$line" if ($dir);
758
759 $$line = "";
760 }
761
762 return $ret;
763 }
764 sub out {
765 my $self = shift;
766 return ($elf ? $self->{value} : undef);
767 }
768 }
769 { package directive; # pick up directives, which start with .
770 sub re {
771 my ($class, $line) = @_;
772 my $self = {};
773 my $ret;
774 my $dir;
775
776 # chain-call to cfi_directive
777 $ret = cfi_directive->re($line) and return $ret;
778
779 if ($$line =~ /^\s*(\.\w+)/) {
780 bless $self,$class;
781 $dir = $1;
782 $ret = $self;
783 undef $self->{value};
784 $$line = substr($$line,@+[0]); $$line =~ s/^\s+//;
785
786 SWITCH: for ($dir) {
787 /\.global|\.globl|\.extern/
788 && do { $globals{$$line} = $prefix . $$line;
789 $$line = $globals{$$line} if ($prefix);
790 last;
791 };
792 /\.type/ && do { my ($sym,$type,$narg) = split(',',$$line);
793 if ($type eq "\@function") {
794 undef $current_function;
795 $current_function->{name} = $sym;
796 $current_function->{abi} = "svr4";
797 $current_function->{narg} = $narg;
798 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";
799 } elsif ($type eq "\@abi-omnipotent") {
800 undef $current_function;
801 $current_function->{name} = $sym;
802 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";
803 }
804 $$line =~ s/\@abi\-omnipotent/\@function/;
805 $$line =~ s/\@function.*/\@function/;
806 last;
807 };
808 /\.asciz/ && do { if ($$line =~ /^"(.*)"$/) {
809 $dir = ".byte";
810 $$line = join(",",unpack("C*",$1),0);
811 }
812 last;
813 };
814 /\.rva|\.long|\.quad/
815 && do { $$line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
816 $$line =~ s/\.L/$decor/g;
817 last;
818 };
819 }
820
821 if ($gas) {
822 $self->{value} = $dir . "\t" . $$line;
823
824 if ($dir =~ /\.extern/) {
825 $self->{value} = ""; # swallow extern
826 } elsif (!$elf && $dir =~ /\.type/) {
827 $self->{value} = "";
828 $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" .
829 (defined($globals{$1})?".scl 2;":".scl 3;") .
830 "\t.type 32;\t.endef"
831 if ($win64 && $$line =~ /([^,]+),\@function/);
832 } elsif (!$elf && $dir =~ /\.size/) {
833 $self->{value} = "";
834 if (defined($current_function)) {
835 $self->{value} .= "${decor}SEH_end_$current_function->{name}:"
836 if ($win64 && $current_function->{abi} eq "svr4");
837 undef $current_function;
838 }
839 } elsif (!$elf && $dir =~ /\.align/) {
840 $self->{value} = ".p2align\t" . (log($$line)/log(2));
841 } elsif ($dir eq ".section") {
842 $current_segment=$$line;
843 if (!$elf && $current_segment eq ".init") {
844 if ($flavour eq "macosx") { $self->{value} = ".mod_init_func"; }
845 elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.ctors"; }
846 }
847 } elsif ($dir =~ /\.(text|data)/) {
848 $current_segment=".$1";
849 } elsif ($dir =~ /\.hidden/) {
850 if ($flavour eq "macosx") { $self->{value} = ".private_extern\t$prefix$$line"; }
851 elsif ($flavour eq "mingw64") { $self->{value} = ""; }
852 } elsif ($dir =~ /\.comm/) {
853 $self->{value} = "$dir\t$prefix$$line";
854 $self->{value} =~ s|,([0-9]+),([0-9]+)$|",$1,".log($2)/log(2)|e if ($flavour eq "macosx");
855 }
856 $$line = "";
857 return $self;
858 }
859
860 # non-gas case or nasm/masm
861 SWITCH: for ($dir) {
862 /\.text/ && do { my $v=undef;
863 if ($nasm) {
864 $v="section .text code align=64\n";
865 } else {
866 $v="$current_segment\tENDS\n" if ($current_segment);
867 $current_segment = ".text\$";
868 $v.="$current_segment\tSEGMENT ";
869 $v.=$masm>=$masmref ? "ALIGN(256)" : "PAGE";
870 $v.=" 'CODE'";
871 }
872 $self->{value} = $v;
873 last;
874 };
875 /\.data/ && do { my $v=undef;
876 if ($nasm) {
877 $v="section .data data align=8\n";
878 } else {
879 $v="$current_segment\tENDS\n" if ($current_segment);
880 $current_segment = "_DATA";
881 $v.="$current_segment\tSEGMENT";
882 }
883 $self->{value} = $v;
884 last;
885 };
886 /\.section/ && do { my $v=undef;
887 $$line =~ s/([^,]*).*/$1/;
888 $$line = ".CRT\$XCU" if ($$line eq ".init");
889 if ($nasm) {
890 $v="section $$line";
891 if ($$line=~/\.([px])data/) {
892 $v.=" rdata align=";
893 $v.=$1 eq "p"? 4 : 8;
894 } elsif ($$line=~/\.CRT\$/i) {
895 $v.=" rdata align=8";
896 }
897 } else {
898 $v="$current_segment\tENDS\n" if ($current_segment);
899 $v.="$$line\tSEGMENT";
900 if ($$line=~/\.([px])data/) {
901 $v.=" READONLY";
902 $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref);
903 } elsif ($$line=~/\.CRT\$/i) {
904 $v.=" READONLY ";
905 $v.=$masm>=$masmref ? "ALIGN(8)" : "DWORD";
906 }
907 }
908 $current_segment = $$line;
909 $self->{value} = $v;
910 last;
911 };
912 /\.extern/ && do { $self->{value} = "EXTERN\t".$$line;
913 $self->{value} .= ":NEAR" if ($masm);
914 last;
915 };
916 /\.globl|.global/
917 && do { $self->{value} = $masm?"PUBLIC":"global";
918 $self->{value} .= "\t".$$line;
919 last;
920 };
921 /\.size/ && do { if (defined($current_function)) {
922 undef $self->{value};
923 if ($current_function->{abi} eq "svr4") {
924 $self->{value}="${decor}SEH_end_$current_function->{name}:";
925 $self->{value}.=":\n" if($masm);
926 }
927 $self->{value}.="$current_function->{name}\tENDP" if($masm && $current_function->{name});
928 undef $current_function;
929 }
930 last;
931 };
932 /\.align/ && do { my $max = ($masm && $masm>=$masmref) ? 256 : 4096;
933 $self->{value} = "ALIGN\t".($$line>$max?$max:$$line);
934 last;
935 };
936 /\.(value|long|rva|quad)/
937 && do { my $sz = substr($1,0,1);
938 my @arr = split(/,\s*/,$$line);
939 my $last = pop(@arr);
940 my $conv = sub { my $var=shift;
941 $var=~s/^(0b[0-1]+)/oct($1)/eig;
942 $var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm);
943 if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva"))
944 { $var=~s/^([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; }
945 $var;
946 };
947
948 $sz =~ tr/bvlrq/BWDDQ/;
949 $self->{value} = "\tD$sz\t";
950 for (@arr) { $self->{value} .= &$conv($_).","; }
951 $self->{value} .= &$conv($last);
952 last;
953 };
954 /\.byte/ && do { my @str=split(/,\s*/,$$line);
955 map(s/(0b[0-1]+)/oct($1)/eig,@str);
956 map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm);
957 while ($#str>15) {
958 $self->{value}.="DB\t"
959 .join(",",@str[0..15])."\n";
960 foreach (0..15) { shift @str; }
961 }
962 $self->{value}.="DB\t"
963 .join(",",@str) if (@str);
964 last;
965 };
966 /\.comm/ && do { my @str=split(/,\s*/,$$line);
967 my $v=undef;
968 if ($nasm) {
969 $v.="common $prefix@str[0] @str[1]";
970 } else {
971 $v="$current_segment\tENDS\n" if ($current_segment);
972 $current_segment = "_DATA";
973 $v.="$current_segment\tSEGMENT\n";
974 $v.="COMM @str[0]:DWORD:".@str[1]/4;
975 }
976 $self->{value} = $v;
977 last;
978 };
979 }
980 $$line = "";
981 }
982
983 $ret;
984 }
985 sub out {
986 my $self = shift;
987 $self->{value};
988 }
989 }
990
991 # Upon initial x86_64 introduction SSE>2 extensions were not introduced
992 # yet. In order not to be bothered by tracing exact assembler versions,
993 # but at the same time to provide a bare security minimum of AES-NI, we
994 # hard-code some instructions. Extensions past AES-NI on the other hand
995 # are traced by examining assembler version in individual perlasm
996 # modules...
997
998 my %regrm = ( "%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3,
999 "%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7 );
1000
1001 sub rex {
1002 my $opcode=shift;
1003 my ($dst,$src,$rex)=@_;
1004
1005 $rex|=0x04 if($dst>=8);
1006 $rex|=0x01 if($src>=8);
1007 push @$opcode,($rex|0x40) if ($rex);
1008 }
1009
1010 my $movq = sub { # elderly gas can't handle inter-register movq
1011 my $arg = shift;
1012 my @opcode=(0x66);
1013 if ($arg =~ /%xmm([0-9]+),\s*%r(\w+)/) {
1014 my ($src,$dst)=($1,$2);
1015 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
1016 rex(\@opcode,$src,$dst,0x8);
1017 push @opcode,0x0f,0x7e;
1018 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M
1019 @opcode;
1020 } elsif ($arg =~ /%r(\w+),\s*%xmm([0-9]+)/) {
1021 my ($src,$dst)=($2,$1);
1022 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
1023 rex(\@opcode,$src,$dst,0x8);
1024 push @opcode,0x0f,0x6e;
1025 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M
1026 @opcode;
1027 } else {
1028 ();
1029 }
1030 };
1031
1032 my $pextrd = sub {
1033 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*(%\w+)/) {
1034 my @opcode=(0x66);
1035 my $imm=$1;
1036 my $src=$2;
1037 my $dst=$3;
1038 if ($dst =~ /%r([0-9]+)d/) { $dst = $1; }
1039 elsif ($dst =~ /%e/) { $dst = $regrm{$dst}; }
1040 rex(\@opcode,$src,$dst);
1041 push @opcode,0x0f,0x3a,0x16;
1042 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M
1043 push @opcode,$imm;
1044 @opcode;
1045 } else {
1046 ();
1047 }
1048 };
1049
1050 my $pinsrd = sub {
1051 if (shift =~ /\$([0-9]+),\s*(%\w+),\s*%xmm([0-9]+)/) {
1052 my @opcode=(0x66);
1053 my $imm=$1;
1054 my $src=$2;
1055 my $dst=$3;
1056 if ($src =~ /%r([0-9]+)/) { $src = $1; }
1057 elsif ($src =~ /%e/) { $src = $regrm{$src}; }
1058 rex(\@opcode,$dst,$src);
1059 push @opcode,0x0f,0x3a,0x22;
1060 push @opcode,0xc0|(($dst&7)<<3)|($src&7); # ModR/M
1061 push @opcode,$imm;
1062 @opcode;
1063 } else {
1064 ();
1065 }
1066 };
1067
1068 my $pshufb = sub {
1069 if (shift =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1070 my @opcode=(0x66);
1071 rex(\@opcode,$2,$1);
1072 push @opcode,0x0f,0x38,0x00;
1073 push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M
1074 @opcode;
1075 } else {
1076 ();
1077 }
1078 };
1079
1080 my $palignr = sub {
1081 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1082 my @opcode=(0x66);
1083 rex(\@opcode,$3,$2);
1084 push @opcode,0x0f,0x3a,0x0f;
1085 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
1086 push @opcode,$1;
1087 @opcode;
1088 } else {
1089 ();
1090 }
1091 };
1092
1093 my $pclmulqdq = sub {
1094 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1095 my @opcode=(0x66);
1096 rex(\@opcode,$3,$2);
1097 push @opcode,0x0f,0x3a,0x44;
1098 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
1099 my $c=$1;
1100 push @opcode,$c=~/^0/?oct($c):$c;
1101 @opcode;
1102 } else {
1103 ();
1104 }
1105 };
1106
1107 my $rdrand = sub {
1108 if (shift =~ /%[er](\w+)/) {
1109 my @opcode=();
1110 my $dst=$1;
1111 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
1112 rex(\@opcode,0,$dst,8);
1113 push @opcode,0x0f,0xc7,0xf0|($dst&7);
1114 @opcode;
1115 } else {
1116 ();
1117 }
1118 };
1119
1120 my $rdseed = sub {
1121 if (shift =~ /%[er](\w+)/) {
1122 my @opcode=();
1123 my $dst=$1;
1124 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
1125 rex(\@opcode,0,$dst,8);
1126 push @opcode,0x0f,0xc7,0xf8|($dst&7);
1127 @opcode;
1128 } else {
1129 ();
1130 }
1131 };
1132
1133 # Not all AVX-capable assemblers recognize AMD XOP extension. Since we
1134 # are using only two instructions hand-code them in order to be excused
1135 # from chasing assembler versions...
1136
1137 sub rxb {
1138 my $opcode=shift;
1139 my ($dst,$src1,$src2,$rxb)=@_;
1140
1141 $rxb|=0x7<<5;
1142 $rxb&=~(0x04<<5) if($dst>=8);
1143 $rxb&=~(0x01<<5) if($src1>=8);
1144 $rxb&=~(0x02<<5) if($src2>=8);
1145 push @$opcode,$rxb;
1146 }
1147
1148 my $vprotd = sub {
1149 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1150 my @opcode=(0x8f);
1151 rxb(\@opcode,$3,$2,-1,0x08);
1152 push @opcode,0x78,0xc2;
1153 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
1154 my $c=$1;
1155 push @opcode,$c=~/^0/?oct($c):$c;
1156 @opcode;
1157 } else {
1158 ();
1159 }
1160 };
1161
1162 my $vprotq = sub {
1163 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1164 my @opcode=(0x8f);
1165 rxb(\@opcode,$3,$2,-1,0x08);
1166 push @opcode,0x78,0xc3;
1167 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
1168 my $c=$1;
1169 push @opcode,$c=~/^0/?oct($c):$c;
1170 @opcode;
1171 } else {
1172 ();
1173 }
1174 };
1175
1176 # Intel Control-flow Enforcement Technology extension. All functions and
1177 # indirect branch targets will have to start with this instruction...
1178
1179 my $endbranch = sub {
1180 (0xf3,0x0f,0x1e,0xfa);
1181 };
1182
1183 ########################################################################
1184
1185 if ($nasm) {
1186 print <<___;
1187 default rel
1188 %define XMMWORD
1189 %define YMMWORD
1190 %define ZMMWORD
1191 ___
1192 } elsif ($masm) {
1193 print <<___;
1194 OPTION DOTNAME
1195 ___
1196 }
1197 while(defined(my $line=<>)) {
1198
1199 $line =~ s|\R$||; # Better chomp
1200
1201 $line =~ s|[#!].*$||; # get rid of asm-style comments...
1202 $line =~ s|/\*.*\*/||; # ... and C-style comments...
1203 $line =~ s|^\s+||; # ... and skip white spaces in beginning
1204 $line =~ s|\s+$||; # ... and at the end
1205
1206 if (my $label=label->re(\$line)) { print $label->out(); }
1207
1208 if (my $directive=directive->re(\$line)) {
1209 printf "%s",$directive->out();
1210 } elsif (my $opcode=opcode->re(\$line)) {
1211 my $asm = eval("\$".$opcode->mnemonic());
1212
1213 if ((ref($asm) eq 'CODE') && scalar(my @bytes=&$asm($line))) {
1214 print $gas?".byte\t":"DB\t",join(',',@bytes),"\n";
1215 next;
1216 }
1217
1218 my @args;
1219 ARGUMENT: while (1) {
1220 my $arg;
1221
1222 ($arg=register->re(\$line, $opcode))||
1223 ($arg=const->re(\$line)) ||
1224 ($arg=ea->re(\$line, $opcode)) ||
1225 ($arg=expr->re(\$line, $opcode)) ||
1226 last ARGUMENT;
1227
1228 push @args,$arg;
1229
1230 last ARGUMENT if ($line !~ /^,/);
1231
1232 $line =~ s/^,\s*//;
1233 } # ARGUMENT:
1234
1235 if ($#args>=0) {
1236 my $insn;
1237 my $sz=$opcode->size();
1238
1239 if ($gas) {
1240 $insn = $opcode->out($#args>=1?$args[$#args]->size():$sz);
1241 @args = map($_->out($sz),@args);
1242 printf "\t%s\t%s",$insn,join(",",@args);
1243 } else {
1244 $insn = $opcode->out();
1245 foreach (@args) {
1246 my $arg = $_->out();
1247 # $insn.=$sz compensates for movq, pinsrw, ...
1248 if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; }
1249 if ($arg =~ /^ymm[0-9]+$/) { $insn.=$sz; $sz="y" if(!$sz); last; }
1250 if ($arg =~ /^zmm[0-9]+$/) { $insn.=$sz; $sz="z" if(!$sz); last; }
1251 if ($arg =~ /^mm[0-9]+$/) { $insn.=$sz; $sz="q" if(!$sz); last; }
1252 }
1253 @args = reverse(@args);
1254 undef $sz if ($nasm && $opcode->mnemonic() eq "lea");
1255 printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args));
1256 }
1257 } else {
1258 printf "\t%s",$opcode->out();
1259 }
1260 }
1261
1262 print $line,"\n";
1263 }
1264
1265 print "$cet_property" if ($cet_property);
1266 print "\n$current_segment\tENDS\n" if ($current_segment && $masm);
1267 print "END\n" if ($masm);
1268
1269 close STDOUT or die "error closing STDOUT: $!;"
1270
1271 \f#################################################
1272 # Cross-reference x86_64 ABI "card"
1273 #
1274 # Unix Win64
1275 # %rax * *
1276 # %rbx - -
1277 # %rcx #4 #1
1278 # %rdx #3 #2
1279 # %rsi #2 -
1280 # %rdi #1 -
1281 # %rbp - -
1282 # %rsp - -
1283 # %r8 #5 #3
1284 # %r9 #6 #4
1285 # %r10 * *
1286 # %r11 * *
1287 # %r12 - -
1288 # %r13 - -
1289 # %r14 - -
1290 # %r15 - -
1291 #
1292 # (*) volatile register
1293 # (-) preserved by callee
1294 # (#) Nth argument, volatile
1295 #
1296 # In Unix terms top of stack is argument transfer area for arguments
1297 # which could not be accommodated in registers. Or in other words 7th
1298 # [integer] argument resides at 8(%rsp) upon function entry point.
1299 # 128 bytes above %rsp constitute a "red zone" which is not touched
1300 # by signal handlers and can be used as temporal storage without
1301 # allocating a frame.
1302 #
1303 # In Win64 terms N*8 bytes on top of stack is argument transfer area,
1304 # which belongs to/can be overwritten by callee. N is the number of
1305 # arguments passed to callee, *but* not less than 4! This means that
1306 # upon function entry point 5th argument resides at 40(%rsp), as well
1307 # as that 32 bytes from 8(%rsp) can always be used as temporal
1308 # storage [without allocating a frame]. One can actually argue that
1309 # one can assume a "red zone" above stack pointer under Win64 as well.
1310 # Point is that at apparently no occasion Windows kernel would alter
1311 # the area above user stack pointer in true asynchronous manner...
1312 #
1313 # All the above means that if assembler programmer adheres to Unix
1314 # register and stack layout, but disregards the "red zone" existence,
1315 # it's possible to use following prologue and epilogue to "gear" from
1316 # Unix to Win64 ABI in leaf functions with not more than 6 arguments.
1317 #
1318 # omnipotent_function:
1319 # ifdef WIN64
1320 # movq %rdi,8(%rsp)
1321 # movq %rsi,16(%rsp)
1322 # movq %rcx,%rdi ; if 1st argument is actually present
1323 # movq %rdx,%rsi ; if 2nd argument is actually ...
1324 # movq %r8,%rdx ; if 3rd argument is ...
1325 # movq %r9,%rcx ; if 4th argument ...
1326 # movq 40(%rsp),%r8 ; if 5th ...
1327 # movq 48(%rsp),%r9 ; if 6th ...
1328 # endif
1329 # ...
1330 # ifdef WIN64
1331 # movq 8(%rsp),%rdi
1332 # movq 16(%rsp),%rsi
1333 # endif
1334 # ret
1335 #
1336 \f#################################################
1337 # Win64 SEH, Structured Exception Handling.
1338 #
1339 # Unlike on Unix systems(*) lack of Win64 stack unwinding information
1340 # has undesired side-effect at run-time: if an exception is raised in
1341 # assembler subroutine such as those in question (basically we're
1342 # referring to segmentation violations caused by malformed input
1343 # parameters), the application is briskly terminated without invoking
1344 # any exception handlers, most notably without generating memory dump
1345 # or any user notification whatsoever. This poses a problem. It's
1346 # possible to address it by registering custom language-specific
1347 # handler that would restore processor context to the state at
1348 # subroutine entry point and return "exception is not handled, keep
1349 # unwinding" code. Writing such handler can be a challenge... But it's
1350 # doable, though requires certain coding convention. Consider following
1351 # snippet:
1352 #
1353 # .type function,@function
1354 # function:
1355 # movq %rsp,%rax # copy rsp to volatile register
1356 # pushq %r15 # save non-volatile registers
1357 # pushq %rbx
1358 # pushq %rbp
1359 # movq %rsp,%r11
1360 # subq %rdi,%r11 # prepare [variable] stack frame
1361 # andq $-64,%r11
1362 # movq %rax,0(%r11) # check for exceptions
1363 # movq %r11,%rsp # allocate [variable] stack frame
1364 # movq %rax,0(%rsp) # save original rsp value
1365 # magic_point:
1366 # ...
1367 # movq 0(%rsp),%rcx # pull original rsp value
1368 # movq -24(%rcx),%rbp # restore non-volatile registers
1369 # movq -16(%rcx),%rbx
1370 # movq -8(%rcx),%r15
1371 # movq %rcx,%rsp # restore original rsp
1372 # magic_epilogue:
1373 # ret
1374 # .size function,.-function
1375 #
1376 # The key is that up to magic_point copy of original rsp value remains
1377 # in chosen volatile register and no non-volatile register, except for
1378 # rsp, is modified. While past magic_point rsp remains constant till
1379 # the very end of the function. In this case custom language-specific
1380 # exception handler would look like this:
1381 #
1382 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1383 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1384 # { ULONG64 *rsp = (ULONG64 *)context->Rax;
1385 # ULONG64 rip = context->Rip;
1386 #
1387 # if (rip >= magic_point)
1388 # { rsp = (ULONG64 *)context->Rsp;
1389 # if (rip < magic_epilogue)
1390 # { rsp = (ULONG64 *)rsp[0];
1391 # context->Rbp = rsp[-3];
1392 # context->Rbx = rsp[-2];
1393 # context->R15 = rsp[-1];
1394 # }
1395 # }
1396 # context->Rsp = (ULONG64)rsp;
1397 # context->Rdi = rsp[1];
1398 # context->Rsi = rsp[2];
1399 #
1400 # memcpy (disp->ContextRecord,context,sizeof(CONTEXT));
1401 # RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase,
1402 # dips->ControlPc,disp->FunctionEntry,disp->ContextRecord,
1403 # &disp->HandlerData,&disp->EstablisherFrame,NULL);
1404 # return ExceptionContinueSearch;
1405 # }
1406 #
1407 # It's appropriate to implement this handler in assembler, directly in
1408 # function's module. In order to do that one has to know members'
1409 # offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant
1410 # values. Here they are:
1411 #
1412 # CONTEXT.Rax 120
1413 # CONTEXT.Rcx 128
1414 # CONTEXT.Rdx 136
1415 # CONTEXT.Rbx 144
1416 # CONTEXT.Rsp 152
1417 # CONTEXT.Rbp 160
1418 # CONTEXT.Rsi 168
1419 # CONTEXT.Rdi 176
1420 # CONTEXT.R8 184
1421 # CONTEXT.R9 192
1422 # CONTEXT.R10 200
1423 # CONTEXT.R11 208
1424 # CONTEXT.R12 216
1425 # CONTEXT.R13 224
1426 # CONTEXT.R14 232
1427 # CONTEXT.R15 240
1428 # CONTEXT.Rip 248
1429 # CONTEXT.Xmm6 512
1430 # sizeof(CONTEXT) 1232
1431 # DISPATCHER_CONTEXT.ControlPc 0
1432 # DISPATCHER_CONTEXT.ImageBase 8
1433 # DISPATCHER_CONTEXT.FunctionEntry 16
1434 # DISPATCHER_CONTEXT.EstablisherFrame 24
1435 # DISPATCHER_CONTEXT.TargetIp 32
1436 # DISPATCHER_CONTEXT.ContextRecord 40
1437 # DISPATCHER_CONTEXT.LanguageHandler 48
1438 # DISPATCHER_CONTEXT.HandlerData 56
1439 # UNW_FLAG_NHANDLER 0
1440 # ExceptionContinueSearch 1
1441 #
1442 # In order to tie the handler to the function one has to compose
1443 # couple of structures: one for .xdata segment and one for .pdata.
1444 #
1445 # UNWIND_INFO structure for .xdata segment would be
1446 #
1447 # function_unwind_info:
1448 # .byte 9,0,0,0
1449 # .rva handler
1450 #
1451 # This structure designates exception handler for a function with
1452 # zero-length prologue, no stack frame or frame register.
1453 #
1454 # To facilitate composing of .pdata structures, auto-generated "gear"
1455 # prologue copies rsp value to rax and denotes next instruction with
1456 # .LSEH_begin_{function_name} label. This essentially defines the SEH
1457 # styling rule mentioned in the beginning. Position of this label is
1458 # chosen in such manner that possible exceptions raised in the "gear"
1459 # prologue would be accounted to caller and unwound from latter's frame.
1460 # End of function is marked with respective .LSEH_end_{function_name}
1461 # label. To summarize, .pdata segment would contain
1462 #
1463 # .rva .LSEH_begin_function
1464 # .rva .LSEH_end_function
1465 # .rva function_unwind_info
1466 #
1467 # Reference to function_unwind_info from .xdata segment is the anchor.
1468 # In case you wonder why references are 32-bit .rvas and not 64-bit
1469 # .quads. References put into these two segments are required to be
1470 # *relative* to the base address of the current binary module, a.k.a.
1471 # image base. No Win64 module, be it .exe or .dll, can be larger than
1472 # 2GB and thus such relative references can be and are accommodated in
1473 # 32 bits.
1474 #
1475 # Having reviewed the example function code, one can argue that "movq
1476 # %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix
1477 # rax would contain an undefined value. If this "offends" you, use
1478 # another register and refrain from modifying rax till magic_point is
1479 # reached, i.e. as if it was a non-volatile register. If more registers
1480 # are required prior [variable] frame setup is completed, note that
1481 # nobody says that you can have only one "magic point." You can
1482 # "liberate" non-volatile registers by denoting last stack off-load
1483 # instruction and reflecting it in finer grade unwind logic in handler.
1484 # After all, isn't it why it's called *language-specific* handler...
1485 #
1486 # SE handlers are also involved in unwinding stack when executable is
1487 # profiled or debugged. Profiling implies additional limitations that
1488 # are too subtle to discuss here. For now it's sufficient to say that
1489 # in order to simplify handlers one should either a) offload original
1490 # %rsp to stack (like discussed above); or b) if you have a register to
1491 # spare for frame pointer, choose volatile one.
1492 #
1493 # (*) Note that we're talking about run-time, not debug-time. Lack of
1494 # unwind information makes debugging hard on both Windows and
1495 # Unix. "Unlike" refers to the fact that on Unix signal handler
1496 # will always be invoked, core dumped and appropriate exit code
1497 # returned to parent (for user notification).
A mirror of the official openssl repository