diff --git a/src/pkg/sync/atomic/asm_386.s b/src/pkg/sync/atomic/asm_386.s
index 0d9dec2167..d149eb66ab 100644
--- a/src/pkg/sync/atomic/asm_386.s
+++ b/src/pkg/sync/atomic/asm_386.s
@@ -103,6 +103,7 @@ TEXT ·LoadInt64(SB),7,$0
 
 TEXT ·LoadUint64(SB),7,$0
 	MOVL	addrptr+0(FP), AX
+	// MOVQ and EMMS were introduced on the Pentium MMX.
 	// MOVQ (%EAX), %MM0
 	BYTE $0x0f; BYTE $0x6f; BYTE $0x00
 	// MOVQ %MM0, 0x8(%ESP)
@@ -131,14 +132,18 @@ TEXT ·StoreInt64(SB),7,$0
 
 TEXT ·StoreUint64(SB),7,$0
 	MOVL	addrptr+0(FP), AX
+	// MOVQ and EMMS were introduced on the Pentium MMX.
 	// MOVQ 0x8(%ESP), %MM0
 	BYTE $0x0f; BYTE $0x6f; BYTE $0x44; BYTE $0x24; BYTE $0x08
 	// MOVQ %MM0, (%EAX)
 	BYTE $0x0f; BYTE $0x7f; BYTE $0x00 
 	// EMMS
 	BYTE $0x0F; BYTE $0x77
-	// MFENCE
-	BYTE $0x0f; BYTE $0xae; BYTE $0xf0
+	// This is essentially a no-op, but it provides required memory fencing.
+	// It can be replaced with MFENCE, but MFENCE was introduced only on the Pentium4 (SSE2).
+	XORL	AX, AX
+	LOCK
+	XADDL	AX, (SP)
 	RET
 
 TEXT ·StoreUintptr(SB),7,$0