Re: [PATCH next 1/3] ipvlan: Defer multicast / broadcast processing to a work-queue
flat view
From: Eric Dumazet <hidden>
Date: 2015-04-24 00:32:32
On Thu, 2015-04-23 at 14:29 -0700, Mahesh Bandewar wrote:
quoted hunk ↗ jump to hunk
+static void ipvlan_multicast_enqueue(struct ipvl_port *port, + struct sk_buff *skb) +{ + if (skb->protocol == htons(ETH_P_PAUSE)) + return;
But what happens to this packet ? It seems leaked ?
quoted hunk ↗ jump to hunk
+ + spin_lock(&port->backlog.lock); + if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) { + __skb_queue_tail(&port->backlog, skb); + spin_unlock(&port->backlog.lock); + } else { + spin_unlock(&port->backlog.lock); + atomic_long_inc(&skb->dev->rx_dropped); + kfree_skb(skb); + } + schedule_work(&port->wq);
No point calling schedule_work(&port->wq); if packet was dropped. We are under pressure, so don't add extra cpu cycles ;)