Virtualization - Aug 2020 - vdpa: handling of VIRTIO_F_ACCESS_PLATFORM/VIRTIO_F_ORDER_PLATFORM

Hi!
I'd like to raise the question of whether we can drop the requirement
of VIRTIO_F_ACCESS_PLATFORM from vdpa?
As far as I can see, it is merely required for virtio vdpa -
so should we not enforce it there?

The point is support for legacy guests - which mostly just works
on x86.

Also, what is the plan for VIRTIO_F_ORDER_PLATFORM?

-- 
MST

On 2020/8/11 ??5:52, Michael S. Tsirkin wrote:
> Hi!
> I'd like to raise the question of whether we can drop the requirement
> of VIRTIO_F_ACCESS_PLATFORM from vdpa?
> As far as I can see, it is merely required for virtio vdpa -
> so should we not enforce it there?

If we don't enforce it, virtio will use PA which breaks the setup when 
IOMMU is enabled. As discussed in the past, mandating DMA API for virito 
can just solve this issue.

>
> The point is support for legacy guests - which mostly just works
> on x86.

Legacy guest should work even if we mandate ACCESS_PLATFORM.

This is because we don't simply pass through guest features (qemu will 
always set ACCESS_PLATFORM to vhost-vdpa).

>
> Also, what is the plan for VIRTIO_F_ORDER_PLATFORM?

I think we should mandate ORDER_PLATFORM, (even for guest).

Thanks
>