[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20221207085941.3b56bc8c@kernel.org>
Date: Wed, 7 Dec 2022 08:59:41 -0800
From: Jakub Kicinski <kuba@...nel.org>
To: Jiri Pirko <jiri@...nulli.us>
Cc: "Kubalewski, Arkadiusz" <arkadiusz.kubalewski@...el.com>,
Vadim Fedorenko <vfedorenko@...ek.ru>,
Jonathan Lemon <jonathan.lemon@...il.com>,
Paolo Abeni <pabeni@...hat.com>,
"netdev@...r.kernel.org" <netdev@...r.kernel.org>,
Vadim Fedorenko <vadfed@...com>,
linux-arm-kernel@...ts.infradead.org, linux-clk@...r.kernel.org,
"Olech, Milena" <milena.olech@...el.com>,
"Michalik, Michal" <michal.michalik@...el.com>
Subject: Re: [RFC PATCH v4 2/4] dpll: Add DPLL framework base functions
On Wed, 7 Dec 2022 14:10:42 +0100 Jiri Pirko wrote:
> >> Why do we need this association at all?
> >
> >Someone someday may want netns delegation and if we don't have the
> >support from the start we may break backward compat introducing it.
>
> Hmm. Can you imagine a usecase?
Running DPLL control in a namespace / container.
I mean - I generally think netns is overused, but yes, it's what
containers use, so I think someone may want to develop their
timer controller SW in as a container?
> Link to devlink instance btw might be a problem. In case of mlx5, one
> dpll instance is going to be created for 2 (or more) PFs. 1 per ConnectX
> ASIC as there is only 1 clock there. And PF devlinks can come and go,
> does not make sense to link it to any of them.
If only we stuck to the "one devlink instance per ASIC", huh? :)
> Thinking about it a bit more, DPLL itself has no network notion. The
> special case is SyncE pin, which is linked to netdevice. Just a small
> part of dpll device. And the netdevice already has notion of netns.
> Isn't that enough?
So we can't use devlink or netdev. Hm. So what do we do?
Make DPLLs only visible in init_net? And require init_net admin?
And when someone comes asking we add an explicit "move to netns"
command to DPLL?
Powered by blists - more mailing lists