diff --git a/arch/sparc/lib/memset.S b/arch/sparc/lib/memset.S
index d79550e3e63f..eaff68213fdf 100644
--- a/arch/sparc/lib/memset.S
+++ b/arch/sparc/lib/memset.S
@@ -27,35 +27,44 @@
 	.text;					\
 	.align	4
 
-#define EXT(start,end,handler) 			\
+#define STORE(source, base, offset, n)		\
+98: 	std source, [base + offset + n];	\
+	.section .fixup,ALLOC,EXECINSTR;	\
+	.align	4;				\
+99:	ba 30f;					\
+	 sub %o3, n - offset, %o3;		\
 	.section __ex_table,ALLOC;		\
 	.align	4;				\
-	.word	start, 0, end, handler;		\
+	.word	98b, 99b;			\
 	.text;					\
-	.align	4
+	.align	4;
+
+#define STORE_LAST(source, base, offset, n)	\
+	EX(std source, [base - offset - n],	\
+	   add %o1, offset + n);
 
 /* Please don't change these macros, unless you change the logic
  * in the .fixup section below as well.
  * Store 64 bytes at (BASE + OFFSET) using value SOURCE. */
-#define ZERO_BIG_BLOCK(base, offset, source)    \
-	std	source, [base + offset + 0x00]; \
-	std	source, [base + offset + 0x08]; \
-	std	source, [base + offset + 0x10]; \
-	std	source, [base + offset + 0x18]; \
-	std	source, [base + offset + 0x20]; \
-	std	source, [base + offset + 0x28]; \
-	std	source, [base + offset + 0x30]; \
-	std	source, [base + offset + 0x38];
+#define ZERO_BIG_BLOCK(base, offset, source)	\
+	STORE(source, base, offset, 0x00);	\
+	STORE(source, base, offset, 0x08);	\
+	STORE(source, base, offset, 0x10);	\
+	STORE(source, base, offset, 0x18);	\
+	STORE(source, base, offset, 0x20);	\
+	STORE(source, base, offset, 0x28);	\
+	STORE(source, base, offset, 0x30);	\
+	STORE(source, base, offset, 0x38);
 
 #define ZERO_LAST_BLOCKS(base, offset, source)	\
-	std	source, [base - offset - 0x38]; \
-	std	source, [base - offset - 0x30]; \
-	std	source, [base - offset - 0x28]; \
-	std	source, [base - offset - 0x20]; \
-	std	source, [base - offset - 0x18]; \
-	std	source, [base - offset - 0x10]; \
-	std	source, [base - offset - 0x08]; \
-	std	source, [base - offset - 0x00];
+	STORE_LAST(source, base, offset, 0x38);	\
+	STORE_LAST(source, base, offset, 0x30); \
+	STORE_LAST(source, base, offset, 0x28);	\
+	STORE_LAST(source, base, offset, 0x20);	\
+	STORE_LAST(source, base, offset, 0x18);	\
+	STORE_LAST(source, base, offset, 0x10);	\
+	STORE_LAST(source, base, offset, 0x08);	\
+	STORE_LAST(source, base, offset, 0x00);
 
 	.text
 	.align 4
@@ -120,8 +129,6 @@ __bzero:
 	ZERO_BIG_BLOCK(%o0, 0x00, %g2)
 	subcc	%o3, 128, %o3
 	ZERO_BIG_BLOCK(%o0, 0x40, %g2)
-11:
-	EXT(10b, 11b, 20f)
 	bne	10b
 	 add	%o0, 128, %o0
 
@@ -136,7 +143,6 @@ __bzero:
 	jmp	%o4
 	 add	%o0, %o2, %o0
 
-12:
 	ZERO_LAST_BLOCKS(%o0, 0x48, %g2)
 	ZERO_LAST_BLOCKS(%o0, 0x08, %g2)
 13:
@@ -182,24 +188,10 @@ __bzero:
 
 	.section .fixup,#alloc,#execinstr
 	.align	4
-20:
-	cmp	%g2, 8
-	bleu	1f
-	 and	%o1, 0x7f, %o1
-	sub	%g2, 9, %g2
-	add	%o3, 64, %o3
-1:
-	sll	%g2, 3, %g2
-	add	%o3, %o1, %o0
+30:
+	and	%o1, 0x7f, %o1
 	retl
-	 sub	%o0, %g2, %o0
-21:
-	mov	8, %o0
-	and	%o1, 7, %o1
-	sub	%o0, %g2, %o0
-	sll	%o0, 3, %o0
-	retl
-	 add	%o0, %o1, %o0
+	 add	%o3, %o1, %o0
 
 	.globl __bzero_end
 __bzero_end: