[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <d0df87e28497a697cae6cd6f03c00d42bc24d764.camel@nvidia.com>
Date: Wed, 15 Dec 2021 18:19:16 +0000
From: Saeed Mahameed <saeedm@...dia.com>
To: "kuba@...nel.org" <kuba@...nel.org>
CC: Parav Pandit <parav@...dia.com>, Jiri Pirko <jiri@...dia.com>,
"netdev@...r.kernel.org" <netdev@...r.kernel.org>,
"davem@...emloft.net" <davem@...emloft.net>,
Sunil Sudhakar Rani <sunrani@...dia.com>,
Bodong Wang <bodong@...dia.com>
Subject: Re: [PATCH net-next 1/2] devlink: Add support to set port function as
trusted
On Thu, 2021-12-02 at 09:31 -0800, Jakub Kicinski wrote:
> On Wed, 1 Dec 2021 07:07:05 +0000 Saeed Mahameed wrote:
> > On Tue, 2021-11-30 at 19:12 -0800, Jakub Kicinski wrote:
> > > On Tue, 30 Nov 2021 22:17:29 +0000 Sunil Sudhakar Rani wrote:
> > > > Sorry for the late response. We agree that the current
> > > > definition
> > > > is vague.
> > > >
> > > > What we meant is that the enforcement is done by device/FW.
> > > > We simply want to allow VF/SF to access privileged or
> > > > restricted
> > > > resource such as physical port counters.
> > > > So how about defining the api such that:
> > > > This knob allows the VF/SF to access restricted resource such
> > > > as
> > > > physical port counters.
> > >
> > > You need to say more about the use case, I don't understand
> > > what you're doing.
> >
> > Some device features/registers/units are not available by default
> > to
> > VFs/SFs (e.g restricted), examples are: physical port
> > registers/counters and similar global attributes.
> >
> > Some customers want to use SF/VF in specialized VM/container for
> > management and monitoring, thus they want SF/VF to have similar
> > privileges to PF in terms of access to restricted resources.
> >
> > Note: this doesn't break the sriov/sf model, trusted SF/VF will not
> > be
> > allowed to alter device attributes, they will simply enjoy access
> > to
> > more resources/features.
>
> None of this explains the use case. It's pretty much what Sunil
> already
> stated.
>
>
After some internal discussions, the plan is to not push new
interfaces, but to utilize the existing devlink params interface for
devlink port functions.
We will suggest a more fine grained parameters to control a port
function (SF/VF) well-defined capabilities.
devlink port function param set/get DEV/PORT_INDEX name PARAMETER value
VALUE cmode { runtime | driverinit | permanent }
Jiri is already on-board. Jakub I hope you are ok with this, let us
know if you have any concerns before we start implementation.
Thanks,
Saeed.
Powered by blists - more mailing lists