Add framework for yet another assembler module dubbed "cpuid." Idea

is to have a placeholder to small routines, which can be written only
in assembler. In IA-32 case this includes processor capability
identification and access to Time-Stamp Counter. As discussed earlier
OPENSSL_ia32cap is introduced to control recently added SSE2 code
pathes (see docs/crypto/OPENSSL_ia32cap.pod). For the moment the
code is operational on ELF platforms only. I haven't checked it yet,
but I have all reasons to believe that Windows build should fail to
link too. I'll be looking into it shortly...

3 files changed, 31 insertions, 3 deletions

diff --git a/crypto/perlasm/x86ms.pl b/crypto/perlasm/x86ms.pl
index dd62348b6a..f6e225c644 100644
--- a/crypto/perlasm/x86ms.pl
+++ b/crypto/perlasm/x86ms.pl
@@ -160,6 +160,8 @@ sub main'jne	{ &out1("jne",@_); }
 sub main'jno	{ &out1("jno",@_); }
 sub main'push	{ &out1("push",@_); $stack+=4; }
 sub main'pop	{ &out1("pop",@_); $stack-=4; }
+sub main'pushf	{ &out0("pushf"); $stack+=4; }
+sub main'popf	{ &out0("popf"); $stack-=4; }
 sub main'bswap	{ &out1("bswap",@_); &using486(); }
 sub main'not	{ &out1("not",@_); }
 sub main'call	{ &out1("call",($_[0]=~/^\$L/?'':'_').$_[0]); }
@@ -168,6 +170,8 @@ sub main'nop	{ &out0("nop"); }
 sub main'test	{ &out2("test",@_); }
 sub main'bt	{ &out2("bt",@_); }
 sub main'leave	{ &out0("leave"); }
+sub main'cpuid	{ &out0("cpuid"); }
+sub main'rdtsc	{ &out0("rdtsc"); }
 
 # SSE2
 sub main'emms	{ &out0("emms"); }
diff --git a/crypto/perlasm/x86nasm.pl b/crypto/perlasm/x86nasm.pl
index da2a1d471b..4cb09ddea6 100644
--- a/crypto/perlasm/x86nasm.pl
+++ b/crypto/perlasm/x86nasm.pl
@@ -169,6 +169,8 @@ sub main'jno	{ &out1("jno NEAR",@_); }
 
 sub main'push	{ &out1("push",@_); $stack+=4; }
 sub main'pop	{ &out1("pop",@_); $stack-=4; }
+sub main'pushf	{ &out0("pushf"); $stack+=4; }
+sub main'popf	{ &out0("popf"); $stack-=4; }
 sub main'bswap	{ &out1("bswap",@_); &using486(); }
 sub main'not	{ &out1("not",@_); }
 sub main'call	{ &out1("call",($_[0]=~/^\$L/?'':'_').$_[0]); }
@@ -177,6 +179,8 @@ sub main'nop	{ &out0("nop"); }
 sub main'test	{ &out2("test",@_); }
 sub main'bt	{ &out2("bt",@_); }
 sub main'leave	{ &out0("leave"); }
+sub main'cpuid	{ &out0("cpuid"); }
+sub main'rdtsc	{ &out0("rdtsc"); }
 
 # SSE2
 sub main'emms	{ &out0("emms"); }
diff --git a/crypto/perlasm/x86unix.pl b/crypto/perlasm/x86unix.pl
index 99e2865aa0..12ff816ebf 100644
--- a/crypto/perlasm/x86unix.pl
+++ b/crypto/perlasm/x86unix.pl
@@ -199,6 +199,8 @@ sub main'nop	{ &out0("nop"); }
 sub main'test	{ &out2("testl",@_); }
 sub main'bt	{ &out2("btl",@_); }
 sub main'leave	{ &out0("leave"); }
+sub main'cpuid	{ &out0(".word\t0xa20f"); }
+sub main'rdtsc	{ &out0(".word\t0x310f"); }
 
 # SSE2
 sub main'emms	{ &out0("emms"); }
@@ -519,11 +521,14 @@ sub main'file_end
 		# 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");
 		$tmp=<<___;
 		cmpl	\$0,(%edx)
 		jne	1f
-		movl	\$1,(%edx)
+		movl	\$1<<10,(%edx)
 		pushf
 		popl	%eax
 		movl	%eax,%ecx
@@ -539,12 +544,13 @@ sub main'file_end
 		pushl	%ebx
 		movl	%edx,%edi
 		movl	\$1,%eax
-		cpuid
-		orl	\$1,%edx
+		.word	0xa20f
+		orl	\$1<<10,%edx
 		movl	%edx,0(%edi)
 		movl	%ecx,4(%edi)
 		popl	%ebx
 		popl	%edi
+	.align	4
 	1:
 ___
 		push (@out,$tmp);
@@ -675,3 +681,17 @@ ___
 	}
 
 sub main'blindpop { &out1("popl",@_); }
+
+sub main'initseg
+	{
+	local($f)=@_;
+	if ($main'elf)
+		{
+		local($tmp)=<<___;
+.pushsection	.init
+	call	$under$f
+.popsection
+___
+		push(@out,$tmp);
+		}
+	}