diff --git a/arch/blackfin/lib/outs.S b/arch/blackfin/lib/outs.S
index 250f4d4b9436..06a5e674401f 100644
--- a/arch/blackfin/lib/outs.S
+++ b/arch/blackfin/lib/outs.S
@@ -13,6 +13,8 @@
 .align 2
 
 ENTRY(_outsl)
+	CC = R2 == 0;
+	IF CC JUMP 1f;
 	P0 = R0;	/* P0 = port */
 	P1 = R1;	/* P1 = address */
 	P2 = R2;	/* P2 = count */
@@ -20,10 +22,12 @@ ENTRY(_outsl)
 	LSETUP( .Llong_loop_s, .Llong_loop_e) LC0 = P2;
 .Llong_loop_s: R0 = [P1++];
 .Llong_loop_e: [P0] = R0;
-	RTS;
+1:	RTS;
 ENDPROC(_outsl)
 
 ENTRY(_outsw)
+	CC = R2 == 0;
+	IF CC JUMP 1f;
 	P0 = R0;	/* P0 = port */
 	P1 = R1;	/* P1 = address */
 	P2 = R2;	/* P2 = count */
@@ -31,10 +35,12 @@ ENTRY(_outsw)
 	LSETUP( .Lword_loop_s, .Lword_loop_e) LC0 = P2;
 .Lword_loop_s: R0 = W[P1++];
 .Lword_loop_e: W[P0] = R0;
-	RTS;
+1:	RTS;
 ENDPROC(_outsw)
 
 ENTRY(_outsb)
+	CC = R2 == 0;
+	IF CC JUMP 1f;
 	P0 = R0;	/* P0 = port */
 	P1 = R1;	/* P1 = address */
 	P2 = R2;	/* P2 = count */
@@ -42,10 +48,12 @@ ENTRY(_outsb)
 	LSETUP( .Lbyte_loop_s, .Lbyte_loop_e) LC0 = P2;
 .Lbyte_loop_s: R0 = B[P1++];
 .Lbyte_loop_e: B[P0] = R0;
-	RTS;
+1:	RTS;
 ENDPROC(_outsb)
 
 ENTRY(_outsw_8)
+	CC = R2 == 0;
+	IF CC JUMP 1f;
 	P0 = R0;	/* P0 = port */
 	P1 = R1;	/* P1 = address */
 	P2 = R2;	/* P2 = count */
@@ -56,5 +64,5 @@ ENTRY(_outsw_8)
 		R0 = R0 << 8;
 		R0 = R0 + R1;
 .Lword8_loop_e: W[P0] = R0;
-	RTS;
+1:	RTS;
 ENDPROC(_outsw_8)
diff --git a/arch/blackfin/mach-common/cache.S b/arch/blackfin/mach-common/cache.S
index 790c767ca95a..ab4a925a443e 100644
--- a/arch/blackfin/mach-common/cache.S
+++ b/arch/blackfin/mach-common/cache.S
@@ -58,6 +58,8 @@
 1:
 .ifeqs "\flushins", BROK_FLUSH_INST
 	\flushins [P0++];
+	nop;
+	nop;
 2:	nop;
 .else
 2:	\flushins [P0++];