Re: [PATCH net-next] netdevsim: Register and unregister devlink traps on probe/remove device

From: Leon Romanovsky
Date: Tue Oct 26 2021 - 03:18:19 EST


On Tue, Oct 26, 2021 at 09:51:13AM +0300, Ido Schimmel wrote:

<...>

> >
> > Can you please explain why is it so important to touch devlink SW
> > objects, reallocate them again and again on every reload in mlxsw?
>
> Because that's how reload was defined and implemented. A complete
> reload. We are not changing the semantics 4 years later.

Please put your emotions aside and explain me technically why are you
must to do it?

The proposed semantics was broken for last 4 years, it can even seen as
dead on arrival, because it never worked for us in real production.

So I'm fixing bugs without relation to when they were introduced.

For example, this fix from Jiri [1] for basic design flow was merged almost
two years later after devlink reload was introduced [2], or this patch from
Parav [3] that fixed an issue introduced year before [4].

[1]
commit 5a508a254bed9a2e36a5fb96c9065532a6bf1e9c
Author: Jiri Pirko <jiri@xxxxxxxxxxxx>
Date: Sat Nov 9 11:29:46 2019 +0100

devlink: disallow reload operation during device cleanup

[2]
commit 2d8dc5bbf4e7603747875eb5cadcd67c1fa8b1bb
Author: Arkadi Sharshevsky <arkadis@xxxxxxxxxxxx>
Date: Mon Jan 15 08:59:04 2018 +0100

devlink: Add support for reload

[3]
commit a7b43649507dae4e55ff0087cad4e4dd1c6d5b99
Author: Parav Pandit <parav@xxxxxxxxxx>
Date: Wed Nov 25 11:16:20 2020 +0200

devlink: Make sure devlink instance and port are in same net namespace

[4]
commit 070c63f20f6c739a3c534555f56c7327536bfcc2
Author: Jiri Pirko <jiri@xxxxxxxxxxxx>
Date: Thu Oct 3 11:49:39 2019 +0200

net: devlink: allow to change namespaces during reload

Thanks