netdev - Re: mlx5e warnings on 6.10

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <80940c9175a29d9bf47d344d195fe6fadec36a98.camel@nvidia.com>
Date: Fri, 19 Jul 2024 16:57:54 +0000
From: Dragos Tatulea <dtatulea@...dia.com>
To: "leitao@...ian.org" <leitao@...ian.org>
CC: Tariq Toukan <tariqt@...dia.com>, Saeed Mahameed <saeedm@...dia.com>,
	"netdev@...r.kernel.org" <netdev@...r.kernel.org>
Subject: Re: mlx5e warnings on 6.10

On Thu, 2024-07-18 at 12:12 -0700, Breno Leitao wrote:
> On Thu, Jul 18, 2024 at 11:00:00AM +0000, Dragos Tatulea wrote:
> > Hi Breno,
> > 
> > On Wed, 2024-07-17 at 04:41 -0700, Breno Leitao wrote:
> 
> > > Sharing in case you find it useful.
> 
> > Thanks for the report. The output, it is very useful. The problem seems to be
> > that mlx5e_tx_reporter_timeout_recover() should take a state lock and doesn't.
> 
> Right. I've looked at other cases where mlx5e_safe_reopen_channels() is
> called, and priv->state_lock is, in fact, hold before calling it.
> 
> So, independent if this fix the problem or not, it seems the right thing
> to do.
> 
> Feel free to add a "Reviewed-by: Breno Leitao <leitao@...ian.org>" when
> you send it.
> 
Thanks!

> > I wonder why this happened only in 6.10. There were no relevant changes in 6.10.
> > Or is it maybe that until now you didn't run into the tx queue timeout issue?
> 
> I don't have a reproducer for it, so, i just got it in 6.10. Maybe just
> a coincidence?
> 
It would be interesting to find out if you are suddenly seeing more tx queue
timeout events than usual in 6.10.

> > Would you have the possibility and willingness to test the below fix?
> 
> Sure. I have two hosts running with your patch, but, it is hard to make
> them timeout.
> 
> Let me know if you have any trick I can explore and force the card to
> time out.
> 
I'd guess that there must be a way. I'm in the best position to find out :).
Once I know I can also test it myself.


Thanks,
Dragos