[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <877ccjzevq.fsf@kurt.kurt.home>
Date: Wed, 14 Aug 2024 10:29:29 +0200
From: Kurt Kanzenbach <kurt@...utronix.de>
To: Maciej Fijalkowski <maciej.fijalkowski@...el.com>, Tony Nguyen
<anthony.l.nguyen@...el.com>
Cc: davem@...emloft.net, kuba@...nel.org, pabeni@...hat.com,
edumazet@...gle.com, netdev@...r.kernel.org, Sriram Yagnaraman
<sriram.yagnaraman@....tech>, magnus.karlsson@...el.com, ast@...nel.org,
daniel@...earbox.net, hawk@...nel.org, john.fastabend@...il.com,
bpf@...r.kernel.org, sriram.yagnaraman@...csson.com,
richardcochran@...il.com, benjamin.steinke@...s-audio.com,
bigeasy@...utronix.de, Chandan Kumar
Rout <chandanx.rout@...el.com>
Subject: Re: [PATCH net-next 3/4] igb: add AF_XDP zero-copy Rx support
>> case XDP_REDIRECT:
>> err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
>
> We were introducing a ZC variant of handling XDP verdict due to a fact
> that mostly what happens is the redirect to user space. We observed a
> reasonable perf improvement from likely()fying XDP_REDIRECT case.
Indeed, I can observe an improvement too.
I've introduced igb_run_xdp_zc() which takes care of that
optimization. Also it takes a pointer to the (read once) xsk_pool to
address the other comment.
Thanks,
Kurt
Download attachment "signature.asc" of type "application/pgp-signature" (862 bytes)
Powered by blists - more mailing lists