Vendor import of OpenSSL 0.9.8i.

4 files changed, 90 insertions, 66 deletions

diff --git a/crypto/perlasm/x86_64-xlate.pl b/crypto/perlasm/x86_64-xlate.pl
index ef1a4ce..a4af769 100755
--- a/crypto/perlasm/x86_64-xlate.pl
+++ b/crypto/perlasm/x86_64-xlate.pl
@@ -57,9 +57,18 @@
 #		lea		.Label-.Lpic_point(%rcx),%rbp
 
 my $output = shift;
-open STDOUT,">$output" || die "can't open $output: $!";
 
-my $masm=1 if ($output =~ /\.asm/);
+{ my ($stddev,$stdino,@junk)=stat(STDOUT);
+  my ($outdev,$outino,@junk)=stat($output);
+
+    open STDOUT,">$output" || die "can't open $output: $!"
+	if ($stddev!=$outdev || $stdino!=$outino);
+}
+
+my $masmref=8 + 50727*2**-32;	# 8.00.50727 shipped with VS2005
+my $masm=$masmref if ($output =~ /\.asm/);
+if ($masm && `ml64 2>&1` =~ m/Version ([0-9]+)\.([0-9]+)(\.([0-9]+))?/)
+{   $masm=$1 + $2*2**-16 + $4*2**-32;   }
 
 my $current_segment;
 my $current_function;
@@ -70,16 +79,18 @@ my $current_function;
 	local	*line = shift;
 	undef	$ret;
 
-	if ($line =~ /^([a-z]+)/i) {
+	if ($line =~ /^([a-z][a-z0-9]*)/i) {
 	    $self->{op} = $1;
 	    $ret = $self;
 	    $line = substr($line,@+[0]); $line =~ s/^\s+//;
 
 	    undef $self->{sz};
-	    if ($self->{op} =~ /(movz)b.*/) {	# movz is pain...
+	    if ($self->{op} =~ /^(movz)b.*/) {	# movz is pain...
 		$self->{op} = $1;
 		$self->{sz} = "b";
-	    } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])/) {
+	    } elsif ($self->{op} =~ /call/) {
+		$self->{sz} = ""
+	    } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) {
 		$self->{op} = $1;
 		$self->{sz} = $2;
 	    }
@@ -95,15 +106,17 @@ my $current_function;
     sub out {
 	my $self = shift;
 	if (!$masm) {
-	    if ($self->{op} eq "movz") {	# movz in pain...
+	    if ($self->{op} eq "movz") {	# movz is pain...
 		sprintf "%s%s%s",$self->{op},$self->{sz},shift;
+	    } elsif ($self->{op} =~ /^set/) { 
+		"$self->{op}";
 	    } elsif ($self->{op} eq "ret") {
 	    	".byte	0xf3,0xc3";
 	    } else {
 		"$self->{op}$self->{sz}";
 	    }
 	} else {
-	    $self->{op} =~ s/movz/movzx/;
+	    $self->{op} =~ s/^movz/movzx/;
 	    if ($self->{op} eq "ret") {
 		$self->{op} = "";
 		if ($current_function->{abi} eq "svr4") {
@@ -133,6 +146,10 @@ my $current_function;
     	my $self = shift;
 
 	if (!$masm) {
+	    # Solaris /usr/ccs/bin/as can't handle multiplications
+	    # in $self->{value}
+	    $self->{value} =~ s/(?<![0-9a-f])(0[x0-9a-f]+)/oct($1)/egi;
+	    $self->{value} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg;
 	    sprintf "\$%s",$self->{value};
 	} else {
 	    $self->{value} =~ s/0x([0-9a-f]+)/0$1h/ig;
@@ -163,14 +180,16 @@ my $current_function;
     	my $self = shift;
 	my $sz = shift;
 
+	# Silently convert all EAs to 64-bit. This is required for
+	# elder GNU assembler and results in more compact code,
+	# *but* most importantly AES module depends on this feature!
+	$self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
+	$self->{base}  =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
+
 	if (!$masm) {
-	    # elder GNU assembler insists on 64-bit EAs:-(
-	    # on pros side, this results in more compact code:-)
-	    $self->{index} =~ s/^[er](.?[0-9xp])[d]?$/r\1/;
-	    $self->{base}  =~ s/^[er](.?[0-9xp])[d]?$/r\1/;
 	    # Solaris /usr/ccs/bin/as can't handle multiplications
 	    # in $self->{label}
-	    $self->{label} =~ s/(?<![0-9a-f])(0[x0-9a-f]+)/oct($1)/eg;
+	    $self->{label} =~ s/(?<![0-9a-f])(0[x0-9a-f]+)/oct($1)/egi;
 	    $self->{label} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg;
 
 	    if (defined($self->{index})) {
@@ -192,6 +211,8 @@ my $current_function;
 					$self->{label},
 					$self->{index},$self->{scale},
 					$self->{base};
+	    } elsif ($self->{base} eq "rip") {
+		sprintf "%s PTR %s",$szmap{$sz},$self->{label};
 	    } else {
 		sprintf "%s PTR %s[%s]",$szmap{$sz},
 					$self->{label},$self->{base};
@@ -317,6 +338,10 @@ my $current_function;
 		$line =~ s/\@function.*/\@function/;
 		if ($line =~ /\.picmeup\s+(%r[\w]+)/i) {
 		    $self->{value} = sprintf "\t.long\t0x%x,0x90000000",$opcode{$1};
+		} elsif ($line =~ /\.asciz\s+"(.*)"$/) {
+		    $self->{value} = ".byte\t".join(",",unpack("C*",$1),0);
+		} elsif ($line =~ /\.extern/) {
+		    $self->{value} = ""; # swallow extern
 		} else {
 		    $self->{value} = $line;
 		}
@@ -334,10 +359,13 @@ my $current_function;
 				    $v="$current_segment\tENDS\n" if ($current_segment);
 				    $current_segment = "_$1\$";
 				    $current_segment =~ tr/[a-z]/[A-Z]/;
-				    $v.="$current_segment\tSEGMENT ALIGN(64) 'CODE'";
+				    $v.="$current_segment\tSEGMENT ";
+				    $v.=$masm>=$masmref ? "ALIGN(64)" : "PAGE";
+				    $v.=" 'CODE'";
 				    $self->{value} = $v;
 				    last;
 				  };
+		/\.extern/  && do { $self->{value} = "EXTRN\t".$line.":BYTE"; last;  };
 		/\.globl/   && do { $self->{value} = "PUBLIC\t".$line; last; };
 		/\.type/    && do { ($sym,$type,$narg) = split(',',$line);
 				    if ($type eq "\@function") {
@@ -362,16 +390,33 @@ my $current_function;
 			    && do { my @arr = split(',',$line);
 				    my $sz  = substr($1,0,1);
 				    my $last = pop(@arr);
+				    my $conv = sub  {	my $var=shift;
+							if ($var=~s/0x([0-9a-f]+)/0$1h/i) { $var; }
+							else { sprintf"0%Xh",$var; }
+						    };  
 
 				    $sz =~ tr/bvlq/BWDQ/;
 				    $self->{value} = "\tD$sz\t";
-				    for (@arr) { $self->{value} .= sprintf"0%Xh,",oct; }
-				    $self->{value} .= sprintf"0%Xh",oct($last);
+				    for (@arr) { $self->{value} .= &$conv($_).","; }
+				    $self->{value} .= &$conv($last);
 				    last;
 				  };
 		/\.picmeup/ && do { $self->{value} = sprintf"\tDD\t 0%Xh,090000000h",$opcode{$line};
 				    last;
 				  };
+		/\.asciz/   && do { if ($line =~ /^"(.*)"$/) {
+					my @str=unpack("C*",$1);
+					push @str,0;
+					while ($#str>15) {
+					    $self->{value}.="DB\t"
+						.join(",",@str[0..15])."\n";
+					    foreach (0..15) { shift @str; }
+					}
+					$self->{value}.="DB\t"
+						.join(",",@str) if (@str);
+				    }
+				    last;
+				  };
 	    }
 	    $line = "";
 	}
@@ -480,7 +525,10 @@ close STDOUT;
 # arguments passed to callee, *but* not less than 4! This means that
 # upon function entry point 5th argument resides at 40(%rsp), as well
 # as that 32 bytes from 8(%rsp) can always be used as temporal
-# storage [without allocating a frame].
+# storage [without allocating a frame]. One can actually argue that
+# one can assume a "red zone" above stack pointer under Win64 as well.
+# Point is that at apparently no occasion Windows kernel would alter
+# the area above user stack pointer in true asynchronous manner...
 #
 # All the above means that if assembler programmer adheres to Unix
 # register and stack layout, but disregards the "red zone" existense,
diff --git a/crypto/perlasm/x86ms.pl b/crypto/perlasm/x86ms.pl
index 82538a9..a0be293 100644
--- a/crypto/perlasm/x86ms.pl
+++ b/crypto/perlasm/x86ms.pl
@@ -146,6 +146,7 @@ sub main'exch	{ &out2("xchg",@_); }
 sub main'cmp	{ &out2("cmp",@_); }
 sub main'lea	{ &out2("lea",@_); }
 sub main'mul	{ &out1("mul",@_); }
+sub main'imul	{ &out2("imul",@_); }
 sub main'div	{ &out1("div",@_); }
 sub main'dec	{ &out1("dec",@_); }
 sub main'inc	{ &out1("inc",@_); }
@@ -204,16 +205,17 @@ sub main'pand	{ &out2("pand",@_); }
 sub out2
 	{
 	local($name,$p1,$p2)=@_;
-	local($l,$t);
+	local($l,$t,$line);
 
-	push(@out,"\t$name\t");
+	$line="\t$name\t";
 	$t=&conv($p1).",";
 	$l=length($t);
-	push(@out,$t);
+	$line.="$t";
 	$l=4-($l+9)/8;
-	push(@out,"\t" x $l);
-	push(@out,&conv($p2));
-	push(@out,"\n");
+	$line.="\t" x $l;
+	$line.=&conv($p2);
+	if ($line=~/\bxmm[0-7]\b/i) { $line=~s/\b[A-Z]+WORD\s+PTR/XMMWORD PTR/i; }
+	push(@out,$line."\n");
 	}
 
 sub out0
@@ -338,11 +340,17 @@ EOF
 sub main'file_end
 	{
 	# try to detect if SSE2 or MMX extensions were used...
-	if (grep {/xmm[0-7]\s*,/i} @out) {
-		grep {s/\.[3-7]86/\.686\n\t\.XMM/} @out;
-		}
-	elsif (grep {/mm[0-7]\s*,/i} @out) {
-		grep {s/\.[3-7]86/\.686\n\t\.MMX/} @out;
+	my $xmmheader=<<___;
+.686
+.XMM
+IF \@Version LT 800
+XMMWORD STRUCT 16
+	DQ  2 dup (?)
+XMMWORD ENDS
+ENDIF
+___
+	if (grep {/\b[x]?mm[0-7]\b/i} @out) {
+		grep {s/\.[3-7]86/$xmmheader/} @out;
 		}
 	push(@out,"_TEXT\$	ENDS\n");
 	push(@out,"END\n");
diff --git a/crypto/perlasm/x86nasm.pl b/crypto/perlasm/x86nasm.pl
index b6dfcbd..fa38f89 100644
--- a/crypto/perlasm/x86nasm.pl
+++ b/crypto/perlasm/x86nasm.pl
@@ -154,6 +154,7 @@ sub main'exch	{ &out2("xchg",@_); }
 sub main'cmp	{ &out2("cmp",@_); }
 sub main'lea	{ &out2("lea",@_); }
 sub main'mul	{ &out1("mul",@_); }
+sub main'imul	{ &out2("imul",@_); }
 sub main'div	{ &out1("div",@_); }
 sub main'dec	{ &out1("dec",@_); }
 sub main'inc	{ &out1("inc",@_); }
@@ -281,6 +282,7 @@ sub main'function_begin
 	my($func,$extra)=@_;
 
 	push(@labels,$func);
+	push(@out,".") if ($main'mwerks);
 	my($tmp)=<<"EOF";
 global	$under$func
 $under$func:
@@ -296,6 +298,7 @@ EOF
 sub main'function_begin_B
 	{
 	my($func,$extra)=@_;
+	push(@out,".") if ($main'mwerks);
 	my($tmp)=<<"EOF";
 global	$under$func
 $under$func:
@@ -375,6 +378,7 @@ sub main'comment
 sub main'public_label
 	{
 	$label{$_[0]}="${under}${_[0]}"	if (!defined($label{$_[0]}));
+	push(@out,".") if ($main'mwerks);
 	push(@out,"global\t$label{$_[0]}\n");
 	}
 
diff --git a/crypto/perlasm/x86unix.pl b/crypto/perlasm/x86unix.pl
index e71050b..a4c9471 100644
--- a/crypto/perlasm/x86unix.pl
+++ b/crypto/perlasm/x86unix.pl
@@ -171,6 +171,7 @@ sub main'exch	{ &out2($_[0]=~/%[a-d][lh]/?"xchgb":"xchgl",@_); }
 sub main'cmp	{ &out2("cmpl",@_); }
 sub main'lea	{ &out2("leal",@_); }
 sub main'mul	{ &out1("mull",@_); }
+sub main'imul	{ &out2("imull",@_); }
 sub main'div	{ &out1("divl",@_); }
 sub main'jmp	{ &out1("jmp",@_); }
 sub main'jmp_ptr { &out1p("jmp",@_); }
@@ -541,50 +542,13 @@ sub main'set_label
 sub main'file_end
 	{
 	# try to detect if SSE2 or MMX extensions were used on ELF platform...
-	if ($main'elf && grep {/%[x]*mm[0-7]/i} @out) {
+	if ($main'elf && grep {/\b%[x]*mm[0-7]\b|OPENSSL_ia32cap_P\b/i} @out) {
 		local($tmp);
 
 		push (@out,"\n.section\t.bss\n");
 		push (@out,".comm\t${under}OPENSSL_ia32cap_P,4,4\n");
 
-		push (@out,".section\t.init\n");
-		# One can argue that it's wasteful to craft every
-		# SSE/MMX module with this snippet... Well, it's 72
-		# bytes long and for the moment we have two modules.
-		# Let's argue when we have 7 modules or so...
-		#
-		# $1<<10 sets a reserved bit to signal that variable
-		# was initialized already...
-		&main'picmeup("edx","OPENSSL_ia32cap_P");
-		$tmp=<<___;
-		cmpl	\$0,(%edx)
-		jne	1f
-		movl	\$1<<10,(%edx)
-		pushf
-		popl	%eax
-		movl	%eax,%ecx
-		xorl	\$1<<21,%eax
-		pushl	%eax
-		popf
-		pushf
-		popl	%eax
-		xorl	%ecx,%eax
-		btl	\$21,%eax
-		jnc	1f
-		pushl	%edi
-		pushl	%ebx
-		movl	%edx,%edi
-		movl	\$1,%eax
-		.byte	0x0f,0xa2
-		orl	\$1<<10,%edx
-		movl	%edx,0(%edi)
-		popl	%ebx
-		popl	%edi
-		jmp	1f
-	.align	$align
-	1:
-___
-		push (@out,$tmp);
+		return;
 	}
 
 	if ($const ne "")