From 287546242539e686ef1233d0a7c57c935fb61c3f Mon Sep 17 00:00:00 2001
From: Andy Polyakov <appro@openssl.org>
Date: Sun, 1 Apr 2007 17:01:12 +0000
Subject: [PATCH] Reserve for assembler implementation of RC4_set_key and
 implement x86 one.

---
 Configure                 |   6 +--
 TABLE                     |  24 ++++-----
 crypto/rc4/Makefile       |   4 +-
 crypto/rc4/asm/rc4-586.pl | 111 +++++++++++++++++++++++++++++++++++---
 4 files changed, 120 insertions(+), 25 deletions(-)

diff --git a/Configure b/Configure
index 89803c5d11..5160b10202 100755
--- a/Configure
+++ b/Configure
@@ -120,8 +120,8 @@ my $x86_elf_asm="x86cpuid-elf.o:bn86-elf.o co86-elf.o mo86-elf.o:dx86-elf.o yx86
 my $x86_coff_asm="x86cpuid-cof.o:bn86-cof.o co86-cof.o mo86-cof.o:dx86-cof.o yx86-cof.o:ax86-cof.o:bx86-cof.o:mx86-cof.o:sx86-cof.o s512sse2-cof.o:cx86-cof.o:rx86-cof.o:rm86-cof.o:r586-cof.o:wp_block.o w86mmx-cof.o";
 my $x86_out_asm="x86cpuid-out.o:bn86-out.o co86-out.o mo86-out.o:dx86-out.o yx86-out.o:ax86-out.o:bx86-out.o:mx86-out.o:sx86-out.o s512sse2-out.o:cx86-out.o:rx86-out.o:rm86-out.o:r586-out.o:wp_block.o w86mmx-out.o";
 
-my $x86_64_asm="x86_64cpuid.o:x86_64-gcc.o x86_64-mont.o::aes-x86_64.o::md5-x86_64.o:sha1-x86_64.o sha256-x86_64.o sha512-x86_64.o::rc4-x86_64.o:::wp-x86_64.o";
-my $ia64_asm="ia64cpuid.o:bn-ia64.o::aes_core.o aes_cbc.o aes-ia64.o::md5-ia64.o:sha1-ia64.o sha256-ia64.o sha512-ia64.o::rc4-ia64.o:::";
+my $x86_64_asm="x86_64cpuid.o:x86_64-gcc.o x86_64-mont.o::aes-x86_64.o::md5-x86_64.o:sha1-x86_64.o sha256-x86_64.o sha512-x86_64.o::rc4-x86_64.o rc4_skey.o:::wp-x86_64.o";
+my $ia64_asm="ia64cpuid.o:bn-ia64.o::aes_core.o aes_cbc.o aes-ia64.o::md5-ia64.o:sha1-ia64.o sha256-ia64.o sha512-ia64.o::rc4-ia64.o rc4_skey.o:::";
 my $sparcv9_asm="sparcv9cap.o sparccpuid.o:bn-sparcv9.o sparcv9-mont.o sparcv9a-mont.o:des_enc-sparc.o fcrypt_b.o:aes_core.o aes_cbc.o aes-sparcv9.o::::::::";
 my $alpha_asm=":bn_asm.o alpha-mont.o::::::::::";
 my $no_asm=":::::::::::";
@@ -599,7 +599,7 @@ my $des_enc="des_enc.o fcrypt_b.o";
 my $aes_enc="aes_core.o aes_cbc.o";
 my $bf_enc	="bf_enc.o";
 my $cast_enc="c_enc.o";
-my $rc4_enc="rc4_enc.o";
+my $rc4_enc="rc4_enc.o rc4_skey.o";
 my $rc5_enc="rc5_enc.o";
 my $md5_obj="";
 my $sha1_obj="";
diff --git a/TABLE b/TABLE
index 18e6e16a97..203cb25e85 100644
--- a/TABLE
+++ b/TABLE
@@ -128,7 +128,7 @@ $bf_obj       =
 $md5_obj      = md5-ia64.o
 $sha1_obj     = sha1-ia64.o sha256-ia64.o sha512-ia64.o
 $cast_obj     = 
-$rc4_obj      = rc4-ia64.o
+$rc4_obj      = rc4-ia64.o rc4_skey.o
 $rmd160_obj   = 
 $rc5_obj      = 
 $wp_obj       = 
@@ -268,7 +268,7 @@ $bf_obj       =
 $md5_obj      = md5-x86_64.o
 $sha1_obj     = sha1-x86_64.o sha256-x86_64.o sha512-x86_64.o
 $cast_obj     = 
-$rc4_obj      = rc4-x86_64.o
+$rc4_obj      = rc4-x86_64.o rc4_skey.o
 $rmd160_obj   = 
 $rc5_obj      = 
 $wp_obj       = wp-x86_64.o
@@ -2172,7 +2172,7 @@ $bf_obj       =
 $md5_obj      = md5-ia64.o
 $sha1_obj     = sha1-ia64.o sha256-ia64.o sha512-ia64.o
 $cast_obj     = 
-$rc4_obj      = rc4-ia64.o
+$rc4_obj      = rc4-ia64.o rc4_skey.o
 $rmd160_obj   = 
 $rc5_obj      = 
 $wp_obj       = 
@@ -2200,7 +2200,7 @@ $bf_obj       =
 $md5_obj      = md5-ia64.o
 $sha1_obj     = sha1-ia64.o sha256-ia64.o sha512-ia64.o
 $cast_obj     = 
-$rc4_obj      = rc4-ia64.o
+$rc4_obj      = rc4-ia64.o rc4_skey.o
 $rmd160_obj   = 
 $rc5_obj      = 
 $wp_obj       = 
@@ -2396,7 +2396,7 @@ $bf_obj       =
 $md5_obj      = md5-ia64.o
 $sha1_obj     = sha1-ia64.o sha256-ia64.o sha512-ia64.o
 $cast_obj     = 
-$rc4_obj      = rc4-ia64.o
+$rc4_obj      = rc4-ia64.o rc4_skey.o
 $rmd160_obj   = 
 $rc5_obj      = 
 $wp_obj       = 
@@ -2424,7 +2424,7 @@ $bf_obj       =
 $md5_obj      = md5-ia64.o
 $sha1_obj     = sha1-ia64.o sha256-ia64.o sha512-ia64.o
 $cast_obj     = 
-$rc4_obj      = rc4-ia64.o
+$rc4_obj      = rc4-ia64.o rc4_skey.o
 $rmd160_obj   = 
 $rc5_obj      = 
 $wp_obj       = 
@@ -2956,7 +2956,7 @@ $bf_obj       =
 $md5_obj      = md5-ia64.o
 $sha1_obj     = sha1-ia64.o sha256-ia64.o sha512-ia64.o
 $cast_obj     = 
-$rc4_obj      = rc4-ia64.o
+$rc4_obj      = rc4-ia64.o rc4_skey.o
 $rmd160_obj   = 
 $rc5_obj      = 
 $wp_obj       = 
@@ -2984,7 +2984,7 @@ $bf_obj       =
 $md5_obj      = md5-ia64.o
 $sha1_obj     = sha1-ia64.o sha256-ia64.o sha512-ia64.o
 $cast_obj     = 
-$rc4_obj      = rc4-ia64.o
+$rc4_obj      = rc4-ia64.o rc4_skey.o
 $rmd160_obj   = 
 $rc5_obj      = 
 $wp_obj       = 
@@ -3012,7 +3012,7 @@ $bf_obj       =
 $md5_obj      = md5-ia64.o
 $sha1_obj     = sha1-ia64.o sha256-ia64.o sha512-ia64.o
 $cast_obj     = 
-$rc4_obj      = rc4-ia64.o
+$rc4_obj      = rc4-ia64.o rc4_skey.o
 $rmd160_obj   = 
 $rc5_obj      = 
 $wp_obj       = 
@@ -3152,7 +3152,7 @@ $bf_obj       =
 $md5_obj      = md5-x86_64.o
 $sha1_obj     = sha1-x86_64.o sha256-x86_64.o sha512-x86_64.o
 $cast_obj     = 
-$rc4_obj      = rc4-x86_64.o
+$rc4_obj      = rc4-x86_64.o rc4_skey.o
 $rmd160_obj   = 
 $rc5_obj      = 
 $wp_obj       = wp-x86_64.o
@@ -3964,7 +3964,7 @@ $bf_obj       =
 $md5_obj      = md5-x86_64.o
 $sha1_obj     = sha1-x86_64.o sha256-x86_64.o sha512-x86_64.o
 $cast_obj     = 
-$rc4_obj      = rc4-x86_64.o
+$rc4_obj      = rc4-x86_64.o rc4_skey.o
 $rmd160_obj   = 
 $rc5_obj      = 
 $wp_obj       = wp-x86_64.o
@@ -3992,7 +3992,7 @@ $bf_obj       =
 $md5_obj      = md5-x86_64.o
 $sha1_obj     = sha1-x86_64.o sha256-x86_64.o sha512-x86_64.o
 $cast_obj     = 
-$rc4_obj      = rc4-x86_64.o
+$rc4_obj      = rc4-x86_64.o rc4_skey.o
 $rmd160_obj   = 
 $rc5_obj      = 
 $wp_obj       = wp-x86_64.o
diff --git a/crypto/rc4/Makefile b/crypto/rc4/Makefile
index 3ecf767682..dcc6d2c816 100644
--- a/crypto/rc4/Makefile
+++ b/crypto/rc4/Makefile
@@ -10,7 +10,7 @@ INCLUDES=
 CFLAG=-g
 AR=		ar r
 
-RC4_ENC=rc4_enc.o
+RC4_ENC=rc4_enc.o rc4_skey.o
 
 CFLAGS= $(INCLUDES) $(CFLAG)
 ASFLAGS= $(INCLUDES) $(ASFLAG)
@@ -22,7 +22,7 @@ APPS=
 
 LIB=$(TOP)/libcrypto.a
 LIBSRC=rc4_skey.c rc4_enc.c
-LIBOBJ=rc4_skey.o $(RC4_ENC)
+LIBOBJ=$(RC4_ENC)
 
 SRC= $(LIBSRC)
 
diff --git a/crypto/rc4/asm/rc4-586.pl b/crypto/rc4/asm/rc4-586.pl
index 22bda4b451..9766abfb6e 100644
--- a/crypto/rc4/asm/rc4-586.pl
+++ b/crypto/rc4/asm/rc4-586.pl
@@ -36,10 +36,6 @@ $in="esi";
 $out="edi";
 $d="ebp";
 
-&RC4("RC4");
-
-&asm_finish();
-
 sub RC4_loop
 	{
 	local($n,$p,$char)=@_;
@@ -99,11 +95,10 @@ sub RC4_loop
 	}
 
 
-sub RC4
+&function_begin_B("RC4");
 	{
 	local($name)=@_;
 
-	&function_begin_B($name,"");
 
 	&mov($ty,&wparam(1));		# len
 	&cmp($ty,0);
@@ -224,7 +219,107 @@ sub RC4
 	 &stack_pop(3);
 	&movb(	&BP(-4,$d,"",0),&LB($y));
 	 &movb(	&BP(-8,$d,"",0),&LB($x));
+}
+&function_end("RC4");
 
-	&function_end($name);
-	}
+########################################################################
+
+$inp="esi";
+$out="edi";
+$idi="ebp";
+$ido="ecx";
+$idx="edx";
+
+&external_label("OPENSSL_ia32cap_P");
+
+# void RC4_set_key(RC4_KEY *key,int len,const unsigned char *data);
+&function_begin("RC4_set_key");
+	&mov	($out,&wparam(0));		# load key
+	&mov	($idi,&wparam(1));		# load len
+	&mov	($inp,&wparam(2));		# load data
+	&picmeup($idx,"OPENSSL_ia32cap_P");
+
+	&lea	($out,&DWP(2*4,$out));		# &key->data
+	&lea	($inp,&DWP(0,$inp,$idi));	# $inp to point at the end
+	&neg	($idi);
+	&xor	("eax","eax");
+	&mov	(&DWP(-4,$out),$idi);		# borrow key->y
+
+	&bt	(&DWP(0,$idx),20);		# check for bit#20
+	&jc	(&label("c1stloop"));
+
+&set_label("w1stloop",16);
+	&mov	(&DWP(0,$out,"eax",4),"eax");	# key->data[i]=i;
+	&add	(&LB("eax"),1);			# i++;
+	&jnc	(&label("w1stloop"));
+
+	&xor	($ido,$ido);
+	&xor	($idx,$idx);
+
+&set_label("w2ndloop",16);
+	&mov	("eax",&DWP(0,$out,$ido,4));
+	&add	(&LB($idx),&BP(0,$inp,$idi));
+	&add	(&LB($idx),&LB("eax"));
+	&add	($idi,1);
+	&mov	("ebx",&DWP(0,$out,$idx,4));
+	&jnz	(&label("wnowrap"));
+	  &mov	($idi,&DWP(-4,$out));
+	&set_label("wnowrap");
+	&mov	(&DWP(0,$out,$idx,4),"eax");
+	&mov	(&DWP(0,$out,$ido,4),"ebx");
+	&add	(&LB($ido),1);
+	&jnc	(&label("w2ndloop"));
+&jmp	(&label("exit"));
+
+&set_label("c1stloop",16);
+	&mov	(&BP(0,$out,"eax"),&LB("eax"));	# key->data[i]=i;
+	&add	(&LB("eax"),1);			# i++;
+	&jnc	(&label("c1stloop"));
+
+	&xor	($ido,$ido);
+	&xor	($idx,$idx);
+	&xor	("ebx","ebx");
+
+&set_label("c2ndloop",16);
+	&mov	(&LB("eax"),&BP(0,$out,$ido));
+	&add	(&LB($idx),&BP(0,$inp,$idi));
+	&add	(&LB($idx),&LB("eax"));
+	&add	($idi,1);
+	&mov	(&LB("ebx"),&BP(0,$out,$idx));
+	&jnz	(&label("cnowrap"));
+	  &mov	($idi,&DWP(-4,$out));
+	&set_label("cnowrap");
+	&mov	(&BP(0,$out,$idx),&LB("eax"));
+	&mov	(&BP(0,$out,$ido),&LB("ebx"));
+	&add	(&LB($ido),1);
+	&jnc	(&label("c2ndloop"));
+
+	&mov	(&DWP(256,$out),-1);		# mark schedule as compressed
+
+&set_label("exit");
+	&xor	("eax","eax");
+	&mov	(&DWP(-8,$out),"eax");		# key->x=0;
+	&mov	(&DWP(-4,$out),"eax");		# key->y=0;
+&function_end("RC4_set_key");
+
+# const char *RC4_options(void);
+&function_begin_B("RC4_options");
+	&call	(&label("pic_point"));
+&set_label("pic_point");
+	&blindpop("eax");
+	&lea	("eax",&DWP(&label("opts")."-".&label("pic_point"),"eax"));
+	&picmeup("edx","OPENSSL_ia32cap_P");
+	&bt	(&DWP(0,"edx"),20);
+	&jnc	(&label("skip"));
+	  &add	("eax",12);
+	&set_label("skip");
+	&ret	();
+&set_label("opts",64);
+&asciz	("rc4(8x,int)");
+&asciz	("rc4(1x,char)");
+&asciz	("RC4 for x86, OpenSSL project");	# RC4_version
+&align	(64);
+&function_end_B("RC4_options");
+
+&asm_finish();
 
-- 
2.25.1