From c067621f5d6d8289d191d8ceea10ba6c6e328ed8 Mon Sep 17 00:00:00 2001
From: tedu <tedu@openbsd.org>
Date: Thu, 9 Jan 2014 05:39:41 +0000
Subject: [PATCH] tiny tweak to asm. prefer memcpy and memmove, with bcopy
 wrapper ok guenther

---
 sys/lib/libkern/arch/amd64/memmove.S | 13 ++++----
 sys/lib/libkern/arch/i386/memmove.S  | 44 +++++++++++++---------------
 2 files changed, 28 insertions(+), 29 deletions(-)

diff --git a/sys/lib/libkern/arch/amd64/memmove.S b/sys/lib/libkern/arch/amd64/memmove.S
index cf81c4c3880..1516c743660 100644
--- a/sys/lib/libkern/arch/amd64/memmove.S
+++ b/sys/lib/libkern/arch/amd64/memmove.S
@@ -40,11 +40,6 @@
 	 *  ws@tools.de     (Wolfgang Solfrank, TooLs GmbH) +49-228-985800
 	 */
 
-ENTRY(memcpy)
-	movq	%rdi,%r11	/* save dest */
-	movq	%rdx,%rcx
-	jmp	2f		/* jump to forward copy code path */
-
 ENTRY(bcopy)
 	xchgq	%rdi,%rsi
 	/* fall into memmove */
@@ -56,7 +51,13 @@ ENTRY(memmove)
 	subq	%rsi,%rax
 	cmpq	%rcx,%rax	/* overlapping? */
 	jb	1f
-2:	cld			/* nope, copy forwards. */
+	jmp	2f		/* nope */
+
+ENTRY(memcpy)
+	movq	%rdi,%r11	/* save dest */
+	movq	%rdx,%rcx
+2:
+	cld			/* copy forwards. */
 	shrq	$3,%rcx		/* copy by words */
 	rep
 	movsq
diff --git a/sys/lib/libkern/arch/i386/memmove.S b/sys/lib/libkern/arch/i386/memmove.S
index 59cadfc01b2..a88b08a5114 100644
--- a/sys/lib/libkern/arch/i386/memmove.S
+++ b/sys/lib/libkern/arch/i386/memmove.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: memmove.S,v 1.5 2013/06/15 19:45:26 miod Exp $	*/
+/*	$OpenBSD: memmove.S,v 1.6 2014/01/09 05:39:41 tedu Exp $	*/
 
 /*-
  * Copyright (c) 1993, 1994, 1995 Charles M. Hannum.  All rights reserved.
@@ -37,46 +37,44 @@
 #include <machine/asm.h>
 
 /*
- * Emulate memmove() by swapping the first two arguments, and jumping
- * into bcopy(), which handles overlapping regions.
+ * Emulate bcopy() by swapping the first two arguments, and jumping
+ * into memmove(), which handles overlapping regions.
  */
-ENTRY(memmove)
+ENTRY(bcopy)
 	pushl	%esi
 	pushl	%edi
-	movl	12(%esp),%edi
-	movl	16(%esp),%esi
+	movl	12(%esp),%esi
+	movl	16(%esp),%edi
 	jmp	docopy
 
 /*
- * Emulate memcpy() by loading the first two arguments in reverse order
- * and jumping into bcopy()'s forward copy code.
+ * memmove(caddr_t dst, caddr_t src, size_t len);
+ * Copy len bytes, coping with overlapping space.
  */
-ENTRY(memcpy)
+ENTRY(memmove)
 	pushl	%esi
 	pushl	%edi
 	movl	12(%esp),%edi
 	movl	16(%esp),%esi
-	movl	20(%esp),%ecx
-	jmp	docopyf
-
-/*
- * bcopy(caddr_t from, caddr_t to, size_t len);
- * Copy len bytes, copying with overlapping space.
- */
-ENTRY(bcopy)
-	pushl	%esi
-	pushl	%edi
-	movl	12(%esp),%esi
-	movl	16(%esp),%edi
 docopy:
 	movl	20(%esp),%ecx
 	movl	%edi,%eax
 	subl	%esi,%eax
 	cmpl	%ecx,%eax		# overlapping?
 	jb	1f
+	jmp	docopyf			# nope
+/*
+ * memcpy() doesn't worry about overlap and always copies forward
+ */
+ENTRY(memcpy)
+	pushl	%esi
+	pushl	%edi
+	movl	12(%esp),%edi
+	movl	16(%esp),%esi
+	movl	20(%esp),%ecx
 docopyf:
 	movl	%edi,%eax		# setup return value for memcpy/memmove
-	cld				# nope, copy forward
+	cld				# copy forward
 	shrl	$2,%ecx			# copy by 32-bit words
 	rep
 	movsl
@@ -89,7 +87,7 @@ docopyf:
 	ret
 
 	_ALIGN_TEXT
-1:	movl	%edi,%eax		# setup return value for memcpy/memmove
+1:	movl	%edi,%eax		# setup return value for memmove
 	addl	%ecx,%edi		# copy backward
 	addl	%ecx,%esi
 	std
-- 
2.20.1