Re: [PATCH v6] net: batch skb dequeueing from softnet input_pkt_queue

[Eric Dumazet <eric.dumazet@gmail.com>]

Le vendredi 23 avril 2010 à 16:12 +0800, Changli Gao a écrit :
> batch skb dequeueing from softnet input_pkt_queue.
> 
> batch skb dequeueing from softnet input_pkt_queue to reduce potential lock
> contention when RPS is enabled.
> 
> Note: in the worst case, the number of packets in a softnet_data may be double
> of netdev_max_backlog.
> 
> Signed-off-by: Changli Gao <xiaosuo@gmail.com>
> ----

Oops, reading it again, I found process_backlog() was still taking the
lock twice, if only one packet is waiting in input_pkt_queue.

Possible fix, on top of your patch :

diff --git a/net/core/dev.c b/net/core/dev.c
index 0eddd23..0569be7 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -3296,8 +3296,9 @@ static int process_backlog(struct napi_struct *napi, int quota)
 #endif
 	napi->weight = weight_p;
 	local_irq_disable();
-	while (1) {
+	while (work < quota) {
 		struct sk_buff *skb;
+		unsigned int qlen;
 
 		while ((skb = __skb_dequeue(&sd->process_queue))) {
 			local_irq_enable();
@@ -3308,13 +3309,15 @@ static int process_backlog(struct napi_struct *napi, int quota)
 		}
 
 		rps_lock(sd);
-		input_queue_head_add(sd, skb_queue_len(&sd->input_pkt_queue));
-		skb_queue_splice_tail_init(&sd->input_pkt_queue,
-					   &sd->process_queue);
-		if (skb_queue_empty(&sd->process_queue)) {
+		qlen = skb_queue_len(&sd->input_pkt_queue);
+		if (qlen) {
+			input_queue_head_add(sd, qlen);
+			skb_queue_splice_tail_init(&sd->input_pkt_queue,
+						   &sd->process_queue);
+		}
+		if (qlen < quota - work) {
 			__napi_complete(napi);
-			rps_unlock(sd);
-			break;
+			quota = work + qlen;
 		}
 		rps_unlock(sd);
 	}