[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20220317111959.7cad8fb1@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com>
Date: Thu, 17 Mar 2022 11:19:59 -0700
From: Jakub Kicinski <kuba@...nel.org>
To: Sriharsha Basavapatna <sriharsha.basavapatna@...adcom.com>
Cc: netdev@...r.kernel.org,
Jiří Pírko <jiri@...nulli.us>,
leonro@...dia.com, saeedm@...dia.com, idosch@...sch.org,
Michael Chan <michael.chan@...adcom.com>,
simon.horman@...igine.com
Subject: Re: [PATCH net-next 1/5] bnxt: use the devlink instance lock to
protect sriov
On Thu, 17 Mar 2022 23:15:33 +0530 Sriharsha Basavapatna wrote:
> The changes look good to me overall. But I have a few concerns. This
> change introduces a lock that is held across modules and if there's
> any upcall from the driver into devlink that might potentially acquire
> the same lock, then it could result in a deadlock. I'm not familiar
> with the internals of devlink, but just want to make sure this point
> is considered. Also, the driver needs to be aware of this lock and use
> it in new code paths within the driver to synchronize with switchdev
> operations. This may not be so obvious when compared to a driver
> private lock.
That's true, that's why we're adding the new "unlocked" devl_* API.
I'm switching the drivers accordingly, I didn't see any upcalls in
the relevant parts in bnxt, LMK if I missed something!
Powered by blists - more mailing lists