Re: [Intel-wired-lan] I225-V (igc driver) hangs after resume in igc_resume/igc_tsn_reset

From: Vinicius Costa Gomes <vinicius.gomes@intel.com>
To: James Hogan <jhogan@kernel.org>,
	Jesse Brandeburg <jesse.brandeburg@intel.com>
Cc: intel-wired-lan <intel-wired-lan@lists.osuosl.org>
Subject: Re: [Intel-wired-lan] I225-V (igc driver) hangs after resume in igc_resume/igc_tsn_reset
Date: Wed, 27 Jul 2022 11:37:09 -0300	[thread overview]
Message-ID: <874jz2ei5m.fsf@intel.com> (raw)
In-Reply-To: <5843458.lOV4Wx5bFT@saruman>

Hi James,

James Hogan <jhogan@kernel.org> writes:

> On Sunday, 17 July 2022 22:40:59 BST James Hogan wrote:
>> On Sunday, 17 July 2022 20:59:36 BST you wrote:
>> > Hi James,
>> > 
>> > James Hogan <jhogan@kernel.org> writes:
>> > > Hi,
>> > > 
>> > > I'm getting regular hangs after resume from suspend with the igc driver,
>> > > for an I225-V (rev 3) on an MSI Pro Z690-A, with version 5.18.11 on
>> > > archlinux. A few stable versions ago it was possible to get the network
>> > > back up by removing and reloading the igc driver, however now I get the
>> > > following, and only a reboot works (which itself hangs before actually
>> > > restarting the machine, and requires a hard reset).
>> > 
>> > Sorry for the delay. I was travelling.
>> 
>> No worries
>> 
>> > I remember seeing some weird behaviors with PCIe PTM and suspend/resume.
>> > Specially with onboard controllers.
>> 
>> It appears that the hardware got itself into a funny state such that
>> NetworkManager hung as described more often than not on resume, however
>> without changing kernel it has now settled back into the previous behaviour
>> of usually working, but occasionally (maybe 1 in 5) the network wouldn't
>> come back up on resume, with network related things hung until I unload and
>> reload the igc module.
>> 
>> > Can you see if disabling CONFIG_PCIE_PTM in your kernel config changes
>> > anything? (assuming it's enabled)
>> 
>> It is enabled yes. Okay I'll give it a go when I get the chance. I'll likely
>> have to do a bunch of boot and suspend cycles to try and get it back into
>> either failure condition.
>
> (sorry somehow dropped others off cc the other day, now adding back)...
>
> I've been running most of this week with 5.18.12-arch1-1, rebuilt with
> CONFIG_PCIE_PTM=n, however I have now observed both cases.
>
> It failed to bring up the network link a couple of times after resume from
> suspend, and i managed to remove the igc module and reload it to get it going
> again.
>
> Another time it failed to come back up, but reloading module didn't help.
>
> I also hit the igc_tsn_reset hang, but this time it was immediately after boot
> (possibly a warm reset), where it failed to bring up the network at all. I'll
> paste the full backtraces of hung tasks below.
>
> I'm wondering whether, since most of the tasks are stuck trying to acquire a
> mutex, the issue is elsewhere. In some past cases though all the tasks that
> are dumped are at a mutex_lock...

Yeah, I agree that it seems like the issue is something else. I would
suggest start with the "simple" things, enabling 'CONFIG_PROVE_LOCKING'
and looking at the first splat, it could be that what you are seeing is
caused by a deadlock somewhere else.

Cheers,
-- 
Vinicius
_______________________________________________
Intel-wired-lan mailing list
Intel-wired-lan@osuosl.org
https://lists.osuosl.org/mailman/listinfo/intel-wired-lan