[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20170106110858.1598bbb8@laptop>
Date: Fri, 6 Jan 2017 11:08:58 -0800
From: Jakub Kicinski <kubakici@...pl>
To: Jiri Pirko <jiri@...nulli.us>
Cc: "Samudrala, Sridhar" <sridhar.samudrala@...el.com>,
alexander.h.duyck@...el.com, john.r.fastabend@...el.com,
anjali.singhai@...el.com, jakub.kicinski@...ronome.com,
davem@...emloft.net, intel-wired-lan@...ts.osuosl.org,
netdev@...r.kernel.org
Subject: Re: [net-next PATCH v2 5/6] i40e: Add TX and RX support in
switchdev mode.
On Fri, 6 Jan 2017 18:30:35 +0100, Jiri Pirko wrote:
> >> > + skb_dst_drop(skb);
> >> > + dst_hold(&priv->vfpr_dst->dst);
> >> > + skb_dst_set(skb, &priv->vfpr_dst->dst);
> >> > + skb->dev = vsi->netdev;
> >> This dst dance seems a bit odd to me. Why don't you just call
> >> i40e_xmit_frame_ring with an extra arg holding the needed metadata?
> >
> >We don't have TX/RX queues associated with VFPR netdevs, so we need to set
> >the dev to PF netdev and requeue the skb.
>
> Still, you eventually call a function within same .c file. Using dst
> does not look right to me.
Do you mean you don't like reusing the dst_metadata to store the
representative id? The missing patch provided the reasoning behind
this design [1]. It's mostly about being able to comfortably use the
queuing infrastructure. Trying to push data from multiple netdevs into
single ring at driver layer is even less pretty.
[1] http://patchwork.ozlabs.org/patch/710563/
Powered by blists - more mailing lists