[dpdk-dev] [PATCH] net/e1000: i350 VLAN tags in loopback packets are big endian
Roger B. Melton
rmelton at cisco.com
Wed Oct 11 16:34:20 CEST 2017
Hi Wenzhou,
On 10/10/17 9:32 PM, Lu, Wenzhuo wrote:
> Hi Roger,
>
>> -----Original Message-----
>> From: Roger B Melton [mailto:rmelton at cisco.com]
>> Sent: Saturday, October 7, 2017 7:34 AM
>> To: Lu, Wenzhuo <wenzhuo.lu at intel.com>
>> Cc: dev at dpdk.org; Roger B Melton <rmelton at cisco.com>
>> Subject: [PATCH] net/e1000: i350 VLAN tags in loopback packets are big
>> endian
>>
>> When copying VLAN tags from the RX descriptor to the vlan_tci field in the
>> mbuf header, igb_rxtx.c:eth_igb_recv_pkts() and
>> eth_igb_recv_scattered_pkts() both assume that the VLAN tag is always little
>> endian. While i350 VLAN non-loopback packets are stored little endian,
>> VLAN tags for i350 VLAN loopback packets are big endian.
>>
>> For i350 VLAN loopback packets, swap the tag when copying from the RX
>> descriptor to the mbuf header. This will ensure that the mbuf vlan_tci is
>> always little endian.
>>
>> Signed-off-by: Roger B Melton <rmelton at cisco.com>
>> ---
>> drivers/net/e1000/igb_rxtx.c | 56
>> ++++++++++++++++++++++++++++++++++++++++----
>> 1 file changed, 51 insertions(+), 5 deletions(-)
>>
>> diff --git a/drivers/net/e1000/igb_rxtx.c b/drivers/net/e1000/igb_rxtx.c index
>> 1c80a2a..8432d8b 100644
>> --- a/drivers/net/e1000/igb_rxtx.c
>> +++ b/drivers/net/e1000/igb_rxtx.c
>
>> @@ -2278,6 +2300,18 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
>>
>> rxq = dev->data->rx_queues[i];
>>
>> + rxq->flags = 0;
>> + /*
>> + * The i210, i211, i350, i354 and i350VF LB vlan packets
>> + * have vlan tags byte swapped.
>> + */
>> + if (hw->mac.type >= e1000_i350) {
> Many thanks for the patch.
> I think we need to check i350 and i354 here as LB bit is only meaningful on these NICs.
Agreed, I'll change this to:
if ((hw->mac.type == e1000_i350) || (hw->mac.type == e1000_i354)) {...
>
>> + rxq->flags |= IGB_RXQ_FLAG_LB_BSWAP_VLAN;
>> + PMD_INIT_LOG(DEBUG, "IGB rx vlan bswap
>> required");
>> + } else {
>> + PMD_INIT_LOG(DEBUG, "IGB rx vlan bswap not
>> required");
>> + }
>> +
>> /* Allocate buffers for descriptor rings and set up queue */
>> ret = igb_alloc_rx_queue_mbufs(rxq);
>> if (ret)
>> @@ -2557,6 +2591,18 @@ eth_igbvf_rx_init(struct rte_eth_dev *dev)
>>
>> rxq = dev->data->rx_queues[i];
>>
>> + rxq->flags = 0;
>> + /*
>> + * The i210, i211, i350, i354 and i350VF LB vlan packets
>> + * have vlan tags byte swapped.
>> + */
>> + if (hw->mac.type >= e1000_i350) {
> We need to check e1000_vfadapt_i350 here to exclude e1000_vfadapt.
Agreed, I'll change this to:
if (hw->mac.type == e1000_vfadapt_i350) {...
I'll make these changes and submit an updated patch.
>
>> + rxq->flags |= IGB_RXQ_FLAG_LB_BSWAP_VLAN;
>> + PMD_INIT_LOG(DEBUG, "IGB rx vlan bswap
>> required");
>> + } else {
>> + PMD_INIT_LOG(DEBUG, "IGB rx vlan bswap not
>> required");
>> + }
>> +
>> /* Allocate buffers for descriptor rings and set up queue */
>> ret = igb_alloc_rx_queue_mbufs(rxq);
>> if (ret)
>> --
>> 2.10.3.dirty
> .
>
More information about the dev
mailing list