Re: [PATCH v5 11/13] KVM: s390: implement mediated device open callback

From: Pierre Morel
Date: Thu Jun 07 2018 - 13:15:24 EST

On 07/06/2018 18:30, Tony Krowiak wrote:
On 06/07/2018 11:20 AM, Pierre Morel wrote:
On 07/06/2018 15:54, Tony Krowiak wrote:
On 06/06/2018 01:40 PM, Pierre Morel wrote:
On 06/06/2018 18:08, Pierre Morel wrote:
On 06/06/2018 16:28, Tony Krowiak wrote:
On 06/05/2018 08:19 AM, Pierre Morel wrote:
On 30/05/2018 16:33, Tony Krowiak wrote:
On 05/24/2018 05:08 AM, Pierre Morel wrote:
On 23/05/2018 16:45, Tony Krowiak wrote:
On 05/16/2018 04:03 AM, Pierre Morel wrote:
On 07/05/2018 17:11, Tony Krowiak wrote:
Implements the open callback on the mediated matrix device.
The function registers a group notifier to receive notification
of the VFIO_GROUP_NOTIFY_SET_KVM event. When notified,
the vfio_ap device driver will get access to the guest's
kvm structure. With access to this structure the driver will:

1. Ensure that only one mediated device is opened for the guest

You should explain why.


2. Configure access to the AP devices for the guest.

...snip...
+void kvm_ap_refcount_inc(struct kvm *kvm)
+{
+ atomic_inc(&kvm->arch.crypto.aprefs);
+}
+EXPORT_SYMBOL(kvm_ap_refcount_inc);
+
+void kvm_ap_refcount_dec(struct kvm *kvm)
+{
+ atomic_dec(&kvm->arch.crypto.aprefs);
+}
+EXPORT_SYMBOL(kvm_ap_refcount_dec);

Why are these functions inside kvm-ap ?
Will anyone use this outer of vfio-ap ?

As I've stated before, I made the choice to contain all interfaces that
access KVM in kvm-ap because I don't think it is appropriate for the device
driver to have to have "knowledge" of the inner workings of KVM. Why does
it matter whether any entity outside of the vfio_ap device driver calls
these functions? I could ask a similar question if the interfaces were
contained in vfio-ap; what if another device driver needs access to these
interfaces?

This is very driver specific and only used during initialization.
It is not a common property of the cryptographic interface.

I really think you should handle this inside the driver.

We are going to have to agree to disagree on this one. Is it not possible
that future drivers - e.g., when full virtualization is implemented - will
require access to KVM?

I do not think that an access to KVM is required for full virtualization.

You may be right, but at this point, there is no guarantee. I stand by my
design on this one.

I really regret that we abandoned the initial design with the matrix bus and one
single parent matrix device per guest.
We would not have the problem of these KVM dependencies.

It had the advantage of taking care of having only one device per guest
(available_instance = 1), could take care of provisioning as you have
sysfs entries available for a matrix without having a guest and a mediated
device.

it also had advantage for virtualization to keep host side and guest side matrix
separate inside parent (host side) and mediated device (guest side).

Shouldn't we treat this problem with a design using standard interfaces
Instead of adding new dedicated interfaces?

Regards,

Pierre



Forget it.

I am not happy with the design but the design I was speaking of may not be the solution either.

The AP architecture makes virtualization of AP devices complex. We tried the solution you
described and found it to be sorely lacking which is why we ended up where we are now.

I did not see any explanation on why between v1 and v2 as it was abandoned.


We have internal structures like the ap_matrix and kvm_ap_matrix
which look like the bus/devices we had previously but are differently
or not at all integrated with the LDD.

What is LDD? Are you talking about dependencies between the vfio_ap device
driver and KVM? If so, see my arguments below.



Also I think that with a little data structure refactoring you can avoid most of
the code in the arch/s390/kvm.

How will structure refactoring help us avoid the code for updating the CRYCB
in the guest's SIE state description.



For example, storing the kvm pointer inside the kvm_ap_matrix and
maintaining a list of the kvm_ap_matrix structures allows to easily know
if a guest already has an associated mediated device.

How is that easier than storing the kvm pointer inside of the mediated matrix
device (i.e., struct ap_matrix_mdev) which also contains the struct kvm_ap_matrix?

you can put it in ap_matrix_mdev but just the name "kvm_ap_matrix" make the
last one a better candidate for my opinion.

How does that allow us to avoid the code in arch/s390/kvm?

This alone does not.

We still need the code
to update the CRYCB in the SIE block. I can obviously avoid placing that code in
kvm-ap.c and move it to the driver, but I already explained my reasoning for
keeping it in KVM. Let's face it, there is no way around the dependency between
the vfio_ap device driver and KVM unless guest matrix configuration is managed
solely by KVM through KVM interfaces.

We get the pointer to KVM from the VFIO interface.
That we both discuss on this is sterile.
The only one who could say what is right is a S390 KVM maintainer.
This would end the discussion.
My point was just to say that we have an alternative.


Why maintain a list of kvm_ap_matrix structures if we don't have to; it is stored
with the mediated matrix device which is passed in to all of the vfio_ap driver
callbacks.

Because using the vm_list which is a static in kvm makes you stick inside the kvm code.


--
Pierre Morel
Linux/KVM/QEMU in BÃblingen - Germany