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

From: Tony Krowiak
Date: Fri Jul 30 2021 - 09:33:18 EST

On 7/28/21 3:42 PM, Halil Pasic wrote:
On Wed, 28 Jul 2021 09:43:03 -0400
Tony Krowiak <akrowiak@xxxxxxxxxxxxx> wrote:

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.
Yes I agree. I was thinking along the lines: matrix modifications
via the sysfs take the matrix_dev->lock so the level of contention
may depend on what userspace is doing...

The probe/remove functions also take the matrix_dev->lock
as does the handle_pqap() function. In any case, while all of
those are possible, in our implementation of AP queue
pass-through, the two functions that take the KVM lock
are invoked when the guest is starting or shutting down,
or when the mdev is hot plugged/unplugged. For the cases of
guest startup/shutdown, it would seem that holding the
kvm->lock while waiting for the matrix_dev->lock shouldn't
be a big problem since the guest will either not be fully up
yet or on its way down. I suppose the hot plug/unplug case
could potentially cause the guest vcpus to pause while processing,
but how often do you anticipate a hot plug to take place?