>-----Original Message----- >From: David Marchand <[email protected]> >Sent: Friday, October 9, 2026 2:20 PM >To: Penigalapati, Sandeep <[email protected]> >Cc: [email protected]; Richardson, Bruce <[email protected]>; >Burakov, Anatoly <[email protected]>; Hore, Soumyadeep ><[email protected]>; [email protected] >Subject: Re: [PATCH] net/ice: fix max number of descriptors for E830 > >My two cents. > >There are still some users of the unconditional macro: >drivers/net/intel/ice/ice_dcf_ethdev.c: .nb_max = ICE_MAX_RING_DESC, >drivers/net/intel/ice/ice_dcf_ethdev.c: .nb_max = ICE_MAX_RING_DESC, >drivers/net/intel/ice/ice_dcf_vf_representor.c: .nb_max = >ICE_MAX_RING_DESC, >drivers/net/intel/ice/ice_dcf_vf_representor.c: .nb_max = >ICE_MAX_RING_DESC, > >Not sure it is a valid problem in DCF case.
Thanks for the review, David. It isn't a valid problem in the DCF case - I verified on E830 (E835) hardware: DCF doesn't advertise RTE_ETH_TX_OFFLOAD_SEND_ON_TIMESTAMP (enabling it is rejected) and a line-rate txonly run showed no MDD. Those DCF users also operate on struct ice_dcf_hw, which has no phy_model, so ICE_MAX_NUM_DESC_BY_MAC() can't apply there anyway. >At least, keeping the ICE_MAX_RING_DESC define seems a good way to >keep/reintroduce similar issues. Fair concern. I didn't change the define since 8160 is still correct for all non-E830 MACs and DCF (only E830 needs 8096) and it's used inside ICE_MAX_NUM_DESC_BY_MAC() itself. v2 adds a comment on it to avoid reintroduction. Regards, Sandeep
