[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20240409144947.1379e33b@kernel.org>
Date: Tue, 9 Apr 2024 14:49:47 -0700
From: Jakub Kicinski <kuba@...nel.org>
To: Ilya Maximets <i.maximets@....org>
Cc: Adrian Moreno <amorenoz@...hat.com>, netdev@...r.kernel.org,
jiri@...nulli.us, xiyou.wangcong@...il.com, cmi@...dia.com,
yotam.gi@...il.com, aconole@...hat.com, echaudro@...hat.com,
horms@...nel.org
Subject: Re: [RFC net-next v2 5/5] net:openvswitch: add psample support
On Tue, 9 Apr 2024 11:35:04 +0200 Ilya Maximets wrote:
> > If we try to implement all our actions following this way, and we keep just
> > copying the incoming actions into the internal representation, we incur in
> > unnecessary memory overhead (e.g: storing 2x struct nlaattr + padding of extra
> > memory to store 2 integers).
> >
> > I don't want to derail the discussion into historical or futuristic changes,
> > just saying that the approach taken in the SAMPLE action (not including this
> > patch) of exposing arguments as attributes but having a kernel-only struct to
> > store them seems to me a good compromise.
>
> Sure. As I said, it's fine to have internal structures. My comment
> was mainly about uAPI part. We should avoid structures in uAPI if
> possible, as they are very hard to maintain and keep compatible with
> older userspace in case some changes will be needed in the future.
FWIW there are some YAML specs for ovs under
Documentation/netlink/specs/ovs_*
perhaps they should also be updated?
Powered by blists - more mailing lists