Re: [RFC blktests fix PATCH] tcp: use GFP_ATOMIC in tcp_disconnect

[Hannes Reinecke <hare@suse.de>]

On 11/25/25 12:54, Nilay Shroff wrote:
> 
> 
> On 11/25/25 5:00 PM, Christoph Hellwig wrote:
>> On Tue, Nov 25, 2025 at 04:58:32PM +0530, Nilay Shroff wrote:
>>> >From git history, I see that was added to avoid memory reclaim  to avoid
>>> possible circular locking dependency. This commit 83e1226b0ee2 ("nvme-tcp:
>>> fix possible circular locking when deleting a controller under memory
>>> pressure") adds it.
>>
>> I suspect this was intended to be noio, and we should just switch to
>> that.
>>
> Yeah, I agree that this should be changed to noio. However, it seems that
> this alone may not be sufficient to fix the lockdep splat reported here.
> Since the real work of fput() might be deferred to another thread, using a noio
> scope in this path may not have the intended effect and could end up being
> redundant.
> 
> That said, I noticed another fix from Chaitanya [1], where fput() is replaced
> with __fput_sync(). With that change in place, combining the noio scope adjustment
> with the switch to __fput_sync() should indeed address the lockdep issue. Given
> that both problems have very similar lockdep signatures, it might make sense to
> merge these two changes into a single fix.
> 
> [1] https://lore.kernel.org/all/20251125005950.41046-1-ckulkarnilinux@gmail.com/
> 
Yes, we should.
Both address similar symptoms, and I wouldn't be surprised both turn out
to address the same issue.
(Frame #0 from here is basically identical to frame #2 in the referenced
issue).

So please roll both into one patch.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke                  Kernel Storage Architect
hare@suse.de                                +49 911 74053 688
SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg
HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich