ppc: fix sungem NAPI (#2)

Hi !

(Resent with proper signed-off line after ack from David)

The recent sungem NAPI change introduced a bug: dev_kfree_skb() is called within the spinlock, thus triggers all sort of WARN_ON's later on down the stack.

This patch changes it to dev_kfree_skb_any(), I hope that is fine as we aren't really in interrupt, are we ? (I don't know in what context NAPI polling occurs, is it a timer IRQ ?)

Ben.

@@ -651,7 +651,7 @@
 		}
 
 		gp->net_stats.tx_packets++;
-		dev_kfree_skb(skb);
+		dev_kfree_skb_any(skb);
 	}
 	gp->tx_old = entry;