[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <Y5BR/n7/rqQ+q8gm@unreal>
Date: Wed, 7 Dec 2022 10:42:38 +0200
From: Leon Romanovsky <leon@...nel.org>
To: Michal Kubecek <mkubecek@...e.cz>
Cc: Jakub Kicinski <kuba@...nel.org>,
Sudheer Mogilappagari <sudheer.mogilappagari@...el.com>,
netdev@...r.kernel.org, andrew@...n.ch, corbet@....net,
sridhar.samudrala@...el.com, anthony.l.nguyen@...el.com
Subject: Re: [PATCH net-next v7] ethtool: add netlink based get rss support
On Tue, Dec 06, 2022 at 05:14:41PM +0100, Michal Kubecek wrote:
> On Mon, Dec 05, 2022 at 09:45:07AM +0200, Leon Romanovsky wrote:
> > On Sun, Dec 04, 2022 at 03:38:50PM -0800, Jakub Kicinski wrote:
> > > Conversion to netlink stands on its own.
> >
> > It doesn't answer on my question. The answer is "we do, just because
> > we can" is nice but doesn't remove my worries that such "future"
> > extension will work with real future feature. From my experience, many
> > UAPI designs without real use case in hand will require adaptions and
> > won't work out-of-box.
> >
> > IMHO, it is the same sin as premature optimization.
>
> Extensibility is likely the most obvious benefit of the netlink
> interface but it's not the only one, even without an immediate need to
> add a new feature, there are other benefits, e.g.
>
> - avoiding the inherently racy get/modify/set cycle
How? IMHO, it is achieved in netlink by holding relevant locks, it can
be rtnl lock or specific to that netlink interface lock (devl). You cam
and should have same locking protection for legacy flow as well.
> - more detailed error reporting thanks to extack
This is extremely good argument.
> - notifications (ethtool --monitor)
>
> And I'm pretty sure the list is not complete. Thus I believe converting
> the ioctl UAPI to netlink is useful even without waiting until we need
> to add new features that would require it.
Thanks
>
> Michal
Powered by blists - more mailing lists