diff --git a/sys/i386/i386/in_cksum.c b/sys/i386/i386/in_cksum.c
index f53f70997be7..d875ed35a912 100644
--- a/sys/i386/i386/in_cksum.c
+++ b/sys/i386/i386/in_cksum.c
@@ -62,15 +62,23 @@
 #define REDUCE          {sum = (sum & 0xffff) + (sum >> 16); ADDCARRY(sum);}
 
 /*
- * Thanks to gcc we don't have to guess
- * which registers contain sum & w.
+ * These asm statements require __volatile because they pass information
+ * via the condition codes.  GCC does not currently provide a way to specify
+ * the condition codes as an input or output operand.
+ *
+ * The LOAD macro below is effectively a prefetch into cache.  GCC will
+ * load the value into a register but will not use it.  Since modern CPUs
+ * reorder operations, this will generally take place in parallel with
+ * other calculations.
  */
 #define ADD(n)	__asm __volatile \
-		("addl " #n "(%1), %0" : "+r" (sum) : "r" (w))
+		("addl %1, %0" : "+r" (sum) : \
+		"g" (((const u_int32_t *)w)[n / 4]))
 #define ADDC(n)	__asm __volatile \
-		("adcl " #n "(%1), %0" : "+r" (sum) : "r" (w))
+		("adcl %1, %0" : "+r" (sum) : \
+		"g" (((const u_int32_t *)w)[n / 4]))
 #define LOAD(n)	__asm __volatile \
-		("movb " #n "(%1), %0" : "=r" (junk) : "r" (w))
+		("" : : "r" (((const u_int32_t *)w)[n / 4]))
 #define MOP	__asm __volatile \
 		("adcl         $0, %0" : "+r" (sum))
 
@@ -163,7 +171,6 @@ in_cksum_skip(m, len, skip)
 		 */
 		mlen -= 1;
 		while ((mlen -= 32) >= 0) {
-			u_char junk;
 			/*
 			 * Add with carry 16 words and fold in the last
 			 * carry by adding a 0 with carry.
diff --git a/sys/i386/include/in_cksum.h b/sys/i386/include/in_cksum.h
index f30c45de79d8..b1cb3f8d98a4 100644
--- a/sys/i386/include/in_cksum.h
+++ b/sys/i386/include/in_cksum.h
@@ -58,10 +58,13 @@ static __inline u_int
 in_cksum_hdr(const struct ip *ip)
 {
 	register u_int sum = 0;
-		    
-#define ADD(n)	__asm("addl " #n "(%1), %0" : "+r" (sum) : "r" (ip))
-#define ADDC(n)	__asm("adcl " #n "(%1), %0" : "+r" (sum) : "r" (ip))
-#define MOP	__asm("adcl         $0, %0" : "+r" (sum))
+
+/* __volatile is necessary here because the condition codes are used. */
+#define ADD(n)	__asm __volatile ("addl %1, %0" : "+r" (sum) : \
+    "g" (((const u_int32_t *)ip)[n / 4]))
+#define ADDC(n)	__asm __volatile ("adcl %1, %0" : "+r" (sum) : \
+    "g" (((const u_int32_t *)ip)[n / 4]))
+#define MOP	__asm __volatile ("adcl $0, %0" : "+r" (sum))
 
 	ADD(0);
 	ADDC(4);
@@ -90,9 +93,9 @@ in_cksum_update(struct ip *ip)
 static __inline u_short
 in_addword(u_short sum, u_short b)
 {
-		    
-	__asm("addw %1, %0" : "+r" (sum) : "r" (b));
-	__asm("adcw $0, %0" : "+r" (sum));
+	/* __volatile is necessary because the condition codes are used. */
+	__asm __volatile ("addw %1, %0" : "+r" (sum) : "r" (b));
+	__asm __volatile ("adcw $0, %0" : "+r" (sum));
 
 	return (sum);
 }
@@ -100,10 +103,10 @@ in_addword(u_short sum, u_short b)
 static __inline u_short
 in_pseudo(u_int sum, u_int b, u_int c)
 {
-		    
-	__asm("addl %1, %0" : "+r" (sum) : "r" (b));
-	__asm("adcl %1, %0" : "+r" (sum) : "r" (c));
-	__asm("adcl $0, %0" : "+r" (sum));
+	/* __volatile is necessary because the condition codes are used. */
+	__asm __volatile ("addl %1, %0" : "+r" (sum) : "g" (b));
+	__asm __volatile ("adcl %1, %0" : "+r" (sum) : "g" (c));
+	__asm __volatile ("adcl $0, %0" : "+r" (sum));
 
 	sum = (sum & 0xffff) + (sum >> 16);
 	if (sum > 0xffff)