From: miod <miod@openbsd.org>
Date: Thu, 17 Apr 2014 21:07:04 +0000 (+0000)
Subject: Remove support for big-endian i386 and amd64.
X-Git-Url: http://artulab.com/gitweb/?a=commitdiff_plain;h=84ba1e83d5755dd6ba8a75641caf8d31c42c122d;p=openbsd

Remove support for big-endian i386 and amd64.

Before someone suggests the OpenSSL people are junkies, here is what they
mention about this:
	/* Most will argue that x86_64 is always little-endian. Well,
	 * yes, but then we have stratus.com who has modified gcc to
	 * "emulate" big-endian on x86. Is there evidence that they
	 * [or somebody else] won't do same for x86_64? Naturally no.
	 * And this line is waiting ready for that brave soul:-) */

So, yes, they are on drugs. But they are not alone, the stratus.com people are,
too.
---

diff --git a/lib/libcrypto/md32_common.h b/lib/libcrypto/md32_common.h
index bfb610e8022..5276abfadca 100644
--- a/lib/libcrypto/md32_common.h
+++ b/lib/libcrypto/md32_common.h
@@ -199,7 +199,6 @@
 # if defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
 #  if ((defined(__i386) || defined(__i386__)) && !defined(I386_ONLY)) || \
       (defined(__x86_64) || defined(__x86_64__))
-#   if !defined(B_ENDIAN)
     /*
      * This gives ~30-40% performance improvement in SHA-256 compiled
      * with gcc [on P4]. Well, first macro to be frank. We can pull
@@ -212,7 +211,6 @@
 #   define HOST_l2c(l,c)	({ unsigned int r=(l);			\
 				   asm ("bswapl %0":"=r"(r):"0"(r));	\
 				   *((unsigned int *)(c))=r; (c)+=4; r;	})
-#   endif
 #  endif
 # endif
 #endif
@@ -251,11 +249,8 @@
 # endif
 #endif
 #if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__)
-# ifndef B_ENDIAN
-/* See comment in DATA_ORDER_IS_BIG_ENDIAN section. */
 #  define HOST_c2l(c,l)	((l)=*((const unsigned int *)(c)), (c)+=4, l)
 #  define HOST_l2c(l,c)	(*((unsigned int *)(c))=(l), (c)+=4, l)
-# endif
 #endif
 
 #ifndef HOST_c2l
diff --git a/lib/libcrypto/sha/sha512.c b/lib/libcrypto/sha/sha512.c
index 50c229ddebf..32bfecbf9bf 100644
--- a/lib/libcrypto/sha/sha512.c
+++ b/lib/libcrypto/sha/sha512.c
@@ -318,13 +318,11 @@ static const SHA_LONG64 K512[80] = {
 				: "=r"(ret)		\
 				: "J"(n),"0"(a)		\
 				: "cc"); ret;		})
-#   if !defined(B_ENDIAN)
 #    define PULL64(x) ({ SHA_LONG64 ret=*((const SHA_LONG64 *)(&(x)));	\
 				asm ("bswapq	%0"		\
 				: "=r"(ret)			\
 				: "0"(ret)); ret;		})
-#   endif
-#  elif (defined(__i386) || defined(__i386__)) && !defined(B_ENDIAN)
+#  elif (defined(__i386) || defined(__i386__))
 #   if defined(I386_ONLY)
 #    define PULL64(x) ({ const unsigned int *p=(const unsigned int *)(&(x));\
 			 unsigned int hi=p[0],lo=p[1];		\
@@ -421,11 +419,7 @@ static void sha512_block_data_order (SHA512_CTX *ctx, const void *in, size_t num
 
 	for (i=0;i<16;i++,F--)
 		{
-#ifdef B_ENDIAN
-		T = W[i];
-#else
 		T = PULL64(W[i]);
-#endif
 		F[0] = A;
 		F[4] = E;
 		F[8] = T;
diff --git a/lib/libcrypto/whrlpool/wp_block.c b/lib/libcrypto/whrlpool/wp_block.c
index 824ed1827c4..ce977083add 100644
--- a/lib/libcrypto/whrlpool/wp_block.c
+++ b/lib/libcrypto/whrlpool/wp_block.c
@@ -84,18 +84,8 @@ typedef unsigned long long	u64;
 #  endif
 #elif defined(__GNUC__) && __GNUC__>=2
 #  if defined(__x86_64) || defined(__x86_64__)
-#    if defined(L_ENDIAN)
 #      define ROTATE(a,n)	({ u64 ret; asm ("rolq %1,%0"	\
 				   : "=r"(ret) : "J"(n),"0"(a) : "cc"); ret; })
-#    elif defined(B_ENDIAN)
-       /* Most will argue that x86_64 is always little-endian. Well,
-        * yes, but then we have stratus.com who has modified gcc to
-	* "emulate" big-endian on x86. Is there evidence that they
-	* [or somebody else] won't do same for x86_64? Naturally no.
-	* And this line is waiting ready for that brave soul:-) */
-#      define ROTATE(a,n)	({ u64 ret; asm ("rorq %1,%0"	\
-				   : "=r"(ret) : "J"(n),"0"(a) : "cc"); ret; })
-#    endif
 #  elif defined(__ia64) || defined(__ia64__)
 #    if defined(L_ENDIAN)
 #      define ROTATE(a,n)	({ u64 ret; asm ("shrp %0=%1,%1,%2"	\
diff --git a/lib/libssl/src/crypto/md32_common.h b/lib/libssl/src/crypto/md32_common.h
index bfb610e8022..5276abfadca 100644
--- a/lib/libssl/src/crypto/md32_common.h
+++ b/lib/libssl/src/crypto/md32_common.h
@@ -199,7 +199,6 @@
 # if defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
 #  if ((defined(__i386) || defined(__i386__)) && !defined(I386_ONLY)) || \
       (defined(__x86_64) || defined(__x86_64__))
-#   if !defined(B_ENDIAN)
     /*
      * This gives ~30-40% performance improvement in SHA-256 compiled
      * with gcc [on P4]. Well, first macro to be frank. We can pull
@@ -212,7 +211,6 @@
 #   define HOST_l2c(l,c)	({ unsigned int r=(l);			\
 				   asm ("bswapl %0":"=r"(r):"0"(r));	\
 				   *((unsigned int *)(c))=r; (c)+=4; r;	})
-#   endif
 #  endif
 # endif
 #endif
@@ -251,11 +249,8 @@
 # endif
 #endif
 #if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__)
-# ifndef B_ENDIAN
-/* See comment in DATA_ORDER_IS_BIG_ENDIAN section. */
 #  define HOST_c2l(c,l)	((l)=*((const unsigned int *)(c)), (c)+=4, l)
 #  define HOST_l2c(l,c)	(*((unsigned int *)(c))=(l), (c)+=4, l)
-# endif
 #endif
 
 #ifndef HOST_c2l
diff --git a/lib/libssl/src/crypto/sha/sha512.c b/lib/libssl/src/crypto/sha/sha512.c
index 50c229ddebf..32bfecbf9bf 100644
--- a/lib/libssl/src/crypto/sha/sha512.c
+++ b/lib/libssl/src/crypto/sha/sha512.c
@@ -318,13 +318,11 @@ static const SHA_LONG64 K512[80] = {
 				: "=r"(ret)		\
 				: "J"(n),"0"(a)		\
 				: "cc"); ret;		})
-#   if !defined(B_ENDIAN)
 #    define PULL64(x) ({ SHA_LONG64 ret=*((const SHA_LONG64 *)(&(x)));	\
 				asm ("bswapq	%0"		\
 				: "=r"(ret)			\
 				: "0"(ret)); ret;		})
-#   endif
-#  elif (defined(__i386) || defined(__i386__)) && !defined(B_ENDIAN)
+#  elif (defined(__i386) || defined(__i386__))
 #   if defined(I386_ONLY)
 #    define PULL64(x) ({ const unsigned int *p=(const unsigned int *)(&(x));\
 			 unsigned int hi=p[0],lo=p[1];		\
@@ -421,11 +419,7 @@ static void sha512_block_data_order (SHA512_CTX *ctx, const void *in, size_t num
 
 	for (i=0;i<16;i++,F--)
 		{
-#ifdef B_ENDIAN
-		T = W[i];
-#else
 		T = PULL64(W[i]);
-#endif
 		F[0] = A;
 		F[4] = E;
 		F[8] = T;
diff --git a/lib/libssl/src/crypto/whrlpool/wp_block.c b/lib/libssl/src/crypto/whrlpool/wp_block.c
index 824ed1827c4..ce977083add 100644
--- a/lib/libssl/src/crypto/whrlpool/wp_block.c
+++ b/lib/libssl/src/crypto/whrlpool/wp_block.c
@@ -84,18 +84,8 @@ typedef unsigned long long	u64;
 #  endif
 #elif defined(__GNUC__) && __GNUC__>=2
 #  if defined(__x86_64) || defined(__x86_64__)
-#    if defined(L_ENDIAN)
 #      define ROTATE(a,n)	({ u64 ret; asm ("rolq %1,%0"	\
 				   : "=r"(ret) : "J"(n),"0"(a) : "cc"); ret; })
-#    elif defined(B_ENDIAN)
-       /* Most will argue that x86_64 is always little-endian. Well,
-        * yes, but then we have stratus.com who has modified gcc to
-	* "emulate" big-endian on x86. Is there evidence that they
-	* [or somebody else] won't do same for x86_64? Naturally no.
-	* And this line is waiting ready for that brave soul:-) */
-#      define ROTATE(a,n)	({ u64 ret; asm ("rorq %1,%0"	\
-				   : "=r"(ret) : "J"(n),"0"(a) : "cc"); ret; })
-#    endif
 #  elif defined(__ia64) || defined(__ia64__)
 #    if defined(L_ENDIAN)
 #      define ROTATE(a,n)	({ u64 ret; asm ("shrp %0=%1,%1,%2"	\