Hi Eric,

On 30 May 10:21 AM, Eric Dumazet wrote:

On Fri, 2014-05-30 at 13:40 -0300, Ezequiel Garcia wrote:

quoted

 
-static int txq_submit_skb(struct tx_queue *txq, struct sk_buff *skb)
+static int txq_submit_skb(struct tx_queue *txq, struct sk_buff *skb,
+			  struct net_device *dev)
 {
 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
 	int nr_frags = skb_shinfo(skb)->nr_frags;

@@ -910,11 +918,15 @@ static int txq_submit_skb(struct tx_queue *txq, struct sk_buff *skb)
 	cmd_sts = 0;
 	l4i_chk = 0;
 
+	if (txq->tx_ring_size - txq->tx_desc_count < MAX_SKB_FRAGS + 1) {

I am not sure I understand this part.

You have one skb here, so why are you using MAX_SKB_FRAGS ?

This check was moved around, so I'm blindly carrying it over.
You meant that I can directly use skb_shinfo(skb)->nr_frags, right?

quoted

+		if (net_ratelimit())
+			netdev_err(dev, "tx queue full?!\n");
+		return -EBUSY;
+	}
+

Also it looks like this part will become dead after the following
patch...

Indeed, I've kept it just for consistency. I had to return some error value
and EBUSY seems the most appropriate. Do you think I should change this?

Thanks for the feedback,
-- 
Ezequiel García, Free Electrons
Embedded Linux, Kernel and Android Engineering
http://free-electrons.com