gro: Optimise length comparison in skb_gro_header

By caching frag0_len, we can avoid checking both frag0 and the
length separately in skb_gro_header.  This helps as skb_gro_header
is called four times per packet which amounts to a few million
times at 10Gb/s.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

include/linux/netdevice.h

@@ -1011,6 +1011,9 @@ struct napi_gro_cb {
 	/* Virtual address of skb_shinfo(skb)->frags[0].page + offset. */
 	void *frag0;
 
+	/* Length of frag0. */
+	unsigned int frag0_len;
+
 	/* This indicates where we are processing relative to skb->data. */
 	int data_offset;
 
@@ -1134,9 +1137,9 @@ static inline void *skb_gro_header(struct sk_buff *skb, unsigned int hlen)
 	unsigned int offset = skb_gro_offset(skb);
 
 	hlen += offset;
-	if (!NAPI_GRO_CB(skb)->frag0 ||
-	    unlikely(skb_shinfo(skb)->frags[0].size < hlen)) {
+	if (NAPI_GRO_CB(skb)->frag0_len < hlen) {
 		NAPI_GRO_CB(skb)->frag0 = NULL;
+		NAPI_GRO_CB(skb)->frag0_len = 0;
 		return pskb_may_pull(skb, hlen) ? skb->data + offset : NULL;
 	}
 

net/core/dev.c

@@ -2509,12 +2509,15 @@ void skb_gro_reset_offset(struct sk_buff *skb)
 {
 	NAPI_GRO_CB(skb)->data_offset = 0;
 	NAPI_GRO_CB(skb)->frag0 = NULL;
+	NAPI_GRO_CB(skb)->frag0_len = 0;
 
 	if (skb->mac_header == skb->tail &&
-	    !PageHighMem(skb_shinfo(skb)->frags[0].page))
+	    !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
 		NAPI_GRO_CB(skb)->frag0 =
 			page_address(skb_shinfo(skb)->frags[0].page) +
 			skb_shinfo(skb)->frags[0].page_offset;
+		NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
+	}
 }
 EXPORT_SYMBOL(skb_gro_reset_offset);