[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20200612212816.GB782829@localhost.localdomain>
Date: Fri, 12 Jun 2020 23:28:16 +0200
From: Lorenzo Bianconi <lorenzo.bianconi@...hat.com>
To: David Miller <davem@...emloft.net>
Cc: lorenzo@...nel.org, netdev@...r.kernel.org,
thomas.petazzoni@...tlin.com, brouer@...hat.com
Subject: Re: [PATCH net] net: mvneta: do not redirect frames during
reconfiguration
> From: Lorenzo Bianconi <lorenzo@...nel.org>
> Date: Tue, 9 Jun 2020 00:02:39 +0200
>
> > Disable frames injection in mvneta_xdp_xmit routine during hw
> > re-configuration in order to avoid hardware hangs
> >
> > Fixes: b0a43db9087a ("net: mvneta: add XDP_TX support")
> > Signed-off-by: Lorenzo Bianconi <lorenzo@...nel.org>
>
> Looking around, I wonder if the fundamental difference from the normal
> TX path is that the XDP path doesn't use the TXQ enable/disable
> machinery and checks like the normal ndo_start_xmit() paths do.
>
> And that's why only the XDP path has this issue.
yes, I agree
>
> I'll apply this, so that the bug is fixed, but note that I consider
> this kind of change adding a new flags mask and one state bit to solve
> a problem to be ultimately inelegant and ususally pointing out a more
> fundamental issue.
I am completely fine to find a common solution since it seems a pattern used
even in other drivers (e.g. bnxt). Reviewing the code probably we need some
checks in __xdp_enqueue() since xdp_ok_fwd_dev() checks just IFF_UP flag.
Regards,
Lorenzo
>
> Thank you.
>
Download attachment "signature.asc" of type "application/pgp-signature" (229 bytes)
Powered by blists - more mailing lists