Re: Lockdep splat on UMS disconnect (6.13-rc1)

[Ming Lei]

On Wed, Dec 04, 2024 at 12:26:53PM +0100, Johan Hovold wrote:
> Hi,
> 
> I'm seeing a new lockdep splat with 6.13-rc1 on resume after having
> disconnected a USB mass storage device while suspended. This does not
> seem to show up with 6.12.
> 
> On one attempt to reproduce this with a ThinkPad T14s, the NVMe was
> hosed after resume as well so I fear that this is not just about missing
> annotation.
> 
> I don't have time to look into this myself right now so I'm reporting in
> the hope that someone familiar with these paths would be able to
> pinpoint any potential regression without too much effort.
> 

The dependency between lock(&q->q_usage_counter(queue)#3) and lock(&q->limits_lock)
is real trouble.

The real deadlock was reported before:

https://lore.kernel.org/linux-scsi/ZxG38G9BuFdBpBHZ@fedora/

And the warning can be triggered by the following command too:

	echo 'none' > /sys/block/sda/queue/write_cache


Thanks,
Ming