[PATCH v2 0/7] net/intel: fixes and improvements to rx path selection
Bruce Richardson
bruce.richardson at intel.com
Thu Oct 16 15:33:20 CEST 2025
On Wed, Oct 15, 2025 at 10:07:16AM +0000, Ciara Loftus wrote:
> This series contains a number of fixes and improvements to the logic
> concerned with selecting an rx path in the intel drivers.
>
> Patch 1 fixes incorrect behaviour in the ci_rxq_vec_capable function
> which would disqualify rxqs that have the timestamp offload enabled from
> selecting vector paths. This was incorrect because iavf vector paths
> support timestamp offload.
>
> Patch 2 removes a reduntant check from the ci_rxq_vec_capable function
> that disqualifies rxqs that have the buffer split offload enabled from
> selecting vector paths. This check is performed during the common rx
> path selection for three of the drivers and during dev_configure for the
> fourth, so the check in ci_rxq_vec_capable is unneeded. Although it is
> similar this patch was kept separate from patch 1 because it is only
> relevant after the three aformention drivers adopted the common rx path
> selection infrastructure, whereas patch 1 is a fix for an issue that
> existed prior to that.
>
> Patches 3 and 4 make fixes to the iavf rx path definitions, ensuring the
> correct offloads and features are defined for all paths.
>
> The final three patches reformat the arrays in the i40e iavf and ice
> drivers that hold information for each Rx path that are used in the
> common selection process, in an attempt to improve readability.
>
> v2:
> * Added a new patch (2) that removes the reduntant check for the buffer
> split offload in the common function.
> * Removed the useless defines IAVF_RX_NO_OFFLOADS and ICE_RX_NO_OFFLOADS
> * Use a newline for closing braces in the rx path infos arrays
> * Let a simd width of zero represent scalar paths instead of 64 which
> lessens the number of fields that need to be initialised for scalar path
> definitions
>
> Ciara Loftus (7):
> net/intel: fix Rx vector capability detection
> net/intel: remove redundant Rx offload check
> net/iavf: fix Rx paths feature definitions
> net/iavf: fix Rx path selection for scalar flex bulk alloc
> net/iavf: reformat the Rx path infos array
> net/i40e: reformat the Rx path infos array
> net/ice: reformat the Rx path infos array
>
Series-acked-by: Bruce Richardson <bruce.richardson at intel.com>
Applied to dpdk-next-net-intel, but with patches 1 & 2 squashed into one. I
don't think that we need two patches to remove the conditional, it's a
basic enough change to do in one.
Thanks,
/Bruce
More information about the dev
mailing list