[dpdk-users] direct specific packets to specific receive queue

Huai-En Tseng the at csie.io
Thu Jul 4 05:54:50 CEST 2019


Hi Stephen,

Thanks for reply.

I have the impression that rte flow raw API only support a few Intel NICs and is only in physical function not virtual function.

Is there other method to match raw byte string and direct to specific receive queue, there are only 2 bytes can be accessed on X520 NIC.

Or should I just buy a Intel 700 series NIC?

By the way, what is flex payload, is this a kind of feature only on Intel 700 series NICs?

Thanks a lot.

> Stephen Hemminger <stephen at networkplumber.org> 於 2019年7月3日 下午10:51 寫道:
> 
> On Wed, 3 Jul 2019 16:05:34 +0800
> Huai-En Tseng <the at csie.io> wrote:
> 
>> Hi, 
>> 
>> I suppose below three things to understand what difference between flex filter, fdir, flex payload and how to use them by tracing source code in testpmd.
>> 
>> First, while I use flex filter API with rte_eth_dev_filter_ctrl(), RTE_ETH_FILTER_FLEXIBLE  and struct rte_eth_flex_filter on my CX4 and X520 its not working.
>> 
>> And I find that in dpdk NIC driver there are only a little NICs like igb NIC support it, so I assume flex filter does not support most NICs such as i40e, ixgbe, ixgbevf and mlx5.
>> 
>> Second, I assume the fdir API is a smaller set which is included the flex filter API.
>> 
>> Third,  flex payload is kind of flex filter API application.
>> 
>> Is all my assumption right?
>> 
>> And is there any full support hardware list?
>> 
>> Last question, is there any method that I can direct specific packets to specific receive queue?
>> 
>> I want to direct pppoe packets to specific receive queue. 
>> 
>> I have tried the raw fdir API but it has limited function in X520, and testpmd flex filter is not working on my CX4 and X520 too.
>> 
>> I also find there is a feature DDP in Intel 700 series NICs but it seems only detects pppoe header but I need to detect ppp header in pppoe header.
>> 
>> Thanks a lot.
> 
> Did you know that flex filter and fdir are older API's that were only supported on a few Intel NIC's.
> 
> 
> Use rte_flow instead.



More information about the users mailing list