DPDK usage discussions
 help / color / mirror / Atom feed
* Flexible payload feature of i40e PMD in DPDK 20.11
@ 2023-01-31 18:44 Joshi, Venkatesh
  2023-01-31 21:40 ` Stephen Hemminger
  0 siblings, 1 reply; 2+ messages in thread
From: Joshi, Venkatesh @ 2023-01-31 18:44 UTC (permalink / raw)
  To: users

Hi,

I have the following configuration:

NIC: Intel XXV710 (supported by i40e PMD)
DPDK version: 20.11

My requirement is to use Flexible payload of the XXV710 NIC (i40e PMD) to do the following:

* For incoming packets where the ethertype is 0x0806 (ie: ARP packets), do the RSS hashing based on the Sender Hardware Address instead of the ethertype field which is considered by default. 
* Basically, I don't want all the ARP packets falling on the same queue. Hence, hashing based on the sender hardware address would give a good distribution across multiple CPUs where my dpdk application is running.


However, I have not been able to find any examples where the following macros have been used - RTE_ETH_INPUT_SET_FLEX_PAYLOAD_1ST_WORD through RTE_ETH_INPUT_SET_FLEX_PAYLOAD_8TH_WORD.


Any pointers would be much appreciated.

Regards,
Venkatesh



^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: Flexible payload feature of i40e PMD in DPDK 20.11
  2023-01-31 18:44 Flexible payload feature of i40e PMD in DPDK 20.11 Joshi, Venkatesh
@ 2023-01-31 21:40 ` Stephen Hemminger
  0 siblings, 0 replies; 2+ messages in thread
From: Stephen Hemminger @ 2023-01-31 21:40 UTC (permalink / raw)
  To: Joshi, Venkatesh; +Cc: users

On Tue, 31 Jan 2023 18:44:51 +0000
"Joshi, Venkatesh" <venkatesh.joshi@hpe.com> wrote:

> Hi,
> 
> I have the following configuration:
> 
> NIC: Intel XXV710 (supported by i40e PMD)
> DPDK version: 20.11
> 
> My requirement is to use Flexible payload of the XXV710 NIC (i40e PMD) to do the following:
> 
> * For incoming packets where the ethertype is 0x0806 (ie: ARP packets), do the RSS hashing based on the Sender Hardware Address instead of the ethertype field which is considered by default. 
> * Basically, I don't want all the ARP packets falling on the same queue. Hence, hashing based on the sender hardware address would give a good distribution across multiple CPUs where my dpdk application is running.
> 
> 
> However, I have not been able to find any examples where the following macros have been used - RTE_ETH_INPUT_SET_FLEX_PAYLOAD_1ST_WORD through RTE_ETH_INPUT_SET_FLEX_PAYLOAD_8TH_WORD.
> 
> 
> Any pointers would be much appreciated.

The old flow direct stuff is deprecated.

What you want could be done via rte_flow API by defining a match for ARP (ethertype)
and action of rte_eth_flow_action_rss of type L2_PAYLOAD and pointing at the sender.

But that assumes the i40e driver implements all this and it doesn't.

You probably have to live will ARP arriving on single queue and do any fanout you want
using rte_ring's.


^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2023-01-31 21:40 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-01-31 18:44 Flexible payload feature of i40e PMD in DPDK 20.11 Joshi, Venkatesh
2023-01-31 21:40 ` Stephen Hemminger

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).