Hi All,
On 09/26/2018 12:17 PM, David Miller wrote:
> From: YueHaibing <yuehaibing@xxxxxxxxxx>
> Date: Wed, 26 Sep 2018 17:09:51 +0800
>
>> @@ -1290,7 +1291,7 @@ static int netcp_ndo_start_xmit(struct sk_buff *skb, struct net_device *ndev)
>> dev_warn(netcp->ndev_dev, "padding failed (%d), packet dropped\n",
>> ret);
>> tx_stats->tx_dropped++;
>> - return ret;
>> + return NETDEV_TX_BUSY;
>> }
>> skb->len = NETCP_MIN_PACKET_SIZE;
>> }
>> @@ -1298,7 +1299,6 @@ static int netcp_ndo_start_xmit(struct sk_buff *skb, struct net_device *ndev)
>> desc = netcp_tx_map_skb(skb, netcp);
>> if (unlikely(!desc)) {
>> netif_stop_subqueue(ndev, subqueue);
>> - ret = -ENOBUFS;
>> goto drop;
>> }
>>
>> @@ -1319,7 +1319,7 @@ static int netcp_ndo_start_xmit(struct sk_buff *skb, struct net_device *ndev)
>> if (desc)
>> netcp_free_tx_desc_chain(netcp, desc, sizeof(*desc));
>> dev_kfree_skb(skb);
>> - return ret;
>> + return NETDEV_TX_BUSY;
>> }
>
> These conversions are not correct.
>
> If the driver frees the SKB you must not return NETDEV_TX_BUSY.
>
> NETDEV_TX_BUSY tells the caller that the driver could not process the
> packet and that it should reqeueu up the SKB and try again later when
> there is more TX queue room.
>
Sry, but I still do not understand the reason for these changes (as I asked already [1])
May be there are some patches on the fly or long term decisions were made in this area.
According to the code include/linux/netdevice.h the .ndo_start_xmit() can return 3 type of values
1) enum netdev_tx, expected to be used by regular netdev drivers, but
2) "error while transmitting (rc < 0)" is also acceptable values
3) NET_XMIT_SUCCESS/DROP/CN (qdisc ->enqueue() return codes) for Virtual network devices
So, are there plans to move NET_XMIT_XXX in enum?
or any patches to change include/linux/netdevice.h "Transmit return codes:" section?
Not sure, that blindly following coccinelle recommendations is a good
choice in this particular case.
[1] https://lkml.org/lkml/2018/9/20/881
--
regards,
-grygorii
