[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20250121090955.25610044@kernel.org>
Date: Tue, 21 Jan 2025 09:09:55 -0800
From: Jakub Kicinski <kuba@...nel.org>
To: Antoine Tenart <atenart@...nel.org>
Cc: davem@...emloft.net, pabeni@...hat.com, edumazet@...gle.com,
stephen@...workplumber.org, gregkh@...uxfoundation.org,
netdev@...r.kernel.org
Subject: Re: [PATCH net-next 3/4] net-sysfs: prevent uncleared queues from
being re-added
On Tue, 21 Jan 2025 10:33:54 +0100 Antoine Tenart wrote:
> Quoting Jakub Kicinski (2025-01-20 20:44:00)
> > On Fri, 17 Jan 2025 11:26:10 +0100 Antoine Tenart wrote:
> > > + if (unlikely(kobj->state_initialized))
> > > + return -EAGAIN;
> >
> > we could do some weird stuff here like try to ignore the sysfs
> > objects and "resynchronize" them before releasing rtnl.
> > Way to hacky to do now, but also debugging a transient EAGAIN
> > will be a major PITA. How about we add a netdev_warn_once()
> > here to leave a trace of what happened in the logs?
>
> I'm not sure as the above can happen in normal conditions, although
> removing and re-adding queues very shortly might be questionable? On the
> other hand I get your point and that might not happen very frequently
> under normal conditions and that's just because I'm hammering the thing
> for testing.
>
> It feel a bit weird to warn something that is not unexpected behavior,
> but if you still prefer having a warn_once for better UX I can add it,
> let me know.
IMHO it's worth adding, but I also don't feel very strongly about it :S
Powered by blists - more mailing lists