Re: [Qemu-devel] [RFC] virtio-mmio: implement modern (v2) personality (virtio-1)

[Laszlo Ersek]

On 08/01/19 01:58, Paolo Bonzini wrote:
> On 30/07/19 18:06, Laszlo Ersek wrote:
>> On 07/29/19 14:57, Sergio Lopez wrote:
>>> Implement the modern (v2) personality, according to the VirtIO 1.0
>>> specification.
>>>
>>> Support for v2 among guests is not as widespread as it'd be
>>> desirable. While the Linux driver has had it for a while, support is
>>> missing, at least, from Tianocore EDK II, NetBSD and FreeBSD.
>>
>> That's right; not only are there no plans to implement virtio-mmio/1.0
>> for OVMF (to my knowledge), I'd even argue against such efforts.
>>
>> OVMF is a heavy-weight guest firmware, which I see entirely out of scope
>> for "micro VMs". And so virtio-mmio/1.0 would seem like a needless &
>> unwelcome complication, from the OVMF maintainership perspective.
> 
> But given that, why not rip out virtio-mmio completely?

Virtio-mmio used to be necessary because "qemu-system-aarch64 -M virt"
lacked a PCI host originally. (The relevant commit is 4ab29b8214cc,
"arm: Add PCIe host bridge in virt machine", 2015-02-13; part of v2.3.0.)

Indeed I don't expect anyone to use virtio-mmio nowadays, and removing
it would simplify both our home-grown VIRTIO_DEVICE_PROTOCOL, and the
virtio drivers.

But it's extra work, not entirely risk-free (regressions), and I can't
tell if someone out there still uses virtio-mmio (despite me thinking
that would be unreasonable). I wouldn't like to see more work sunk into
it either way :)

Laszlo