Re: [PATCH 2/2] s390/vfio-ap: replace open coded locks for VFIO_GROUP_NOTIFY_SET_KVM notification

[Tony Krowiak <akrowiak@linux.ibm.com>]

On 7/26/21 6:43 PM, Halil Pasic wrote:
> On Mon, 26 Jul 2021 19:03:17 -0300
> Jason Gunthorpe <jgg@nvidia.com> wrote:
>
>>> You may end up with open and close running interleaved. What I'
>>> trying to say is, to my best knowledge, normally there is no
>>> you have to close it before you open it again rule for files.
>> This is an existing bug in this driver, I've fixed in the reflck series.
>>
>> open_device/close_device will not run concurrently, or out of order,
>> afer it is fixed.
> Well if that is the case then provided your fix precedes this patch:
>
> Acked-by: Halil Pasic <pasic@linux.ibm.com>
>
> I'm not entirely happy with this. I did not thoroughly investigate the
> implications of reversing the locking order of the vfio-ap lock (driver
> global) and the kvm lock (guest specific). I will trust Tony and hope
> our KVM maintainers will scream if this is bad from interference and
> delay perspective.

This solution was suggested by Jason G and it does in fact resolve
the lockdep splat encountered when starting an SE guest with
access to crypto resources. There is a chance that the KVM lock
can get held while waiting for the lock on the matrix_dev->mutex,
but this does not seem like a grave concern to me. That would
only happen during VFIO_GROUP_NOTIFY_SET_KVM notification - either
when the guest is being started or terminated, or when the mdev
fd is opened or closed. According to Jason, once he integrates his reflck
series, the open/close will happen only once.

I've copied the KVM maintainers on this response, so hopefully one of
them will provide some input.

>
> Regards,
> Halil