RE: [virtio-comment] [PATCH v9 0/4] admin: Access legacy registers using admin commands

[Cornelia Huck <cohuck@redhat.com>]

On Wed, Jul 05 2023, Parav Pandit <parav@nvidia.com> wrote:

>> From: Michael S. Tsirkin <mst@redhat.com>
>> Sent: Wednesday, July 5, 2023 7:09 AM
>> 
>> OK I sent a bunch of comments. Pls try to address, unfortunately I didn't go in
>> depth over the last patch, but I feel it should be squashed into the generic
>> legacy command sections avoiding need for indirection, so rewriting it now
>> wouldn't be efficient.
>
> I relooked at well yday.
> Transport has very small section.
> This aligns to how the existing spec is written, for example vq notify data is explained in generic way and defined in pci way in PCI chapter.
> The indirection is fine and also self-contained similarly here.

Would it make sense to do this the other way around (i.e. use something
that is PCI-only from the start?)

IIUC, the "legacy" problem as it stands is something specific to PCI,
with MMIO using two distinct approaches for pre-1.0 and 1.0+, and CCW
simply using different commands (and no need for any kind of cross
access.) Not 100% sure about MMIO, but I don't think we need a version
of this for CCW.