Re: [PATCH v2] hw/acpi: Add vmclock device

[David Woodhouse]

On 30 July 2024 21:45:53 BST, "Michael S. Tsirkin" <mst@redhat.com> wrote:
>On Tue, Jul 30, 2024 at 08:04:17PM +0100, David Woodhouse wrote:
>> On 30 July 2024 18:53:18 BST, "Michael S. Tsirkin" <mst@redhat.com> wrote:
>> >We don't want to manually sync headers with Linux.
>> 
>> Indeed. I was briefly tempted to fake it, but figured it might get lost if 
>> we subsequently do run the script to automatically merge from Linux, before 
>> the guest driver is merged there.
>> 
>> >I think Linux abi should live under uapi. When it is there, we can use
>> >./scripts/update-linux-headers.sh machinery to import it.
>> 
>> This isn't just Linux ABI. It's intended as hypervisor to guest ABI too. In 
>> the fullness of time I'm hoping it'll actually be a virtio header. In the 
>> meantime, best not to overthink it. It's fine in hw/acpi alongside the 
>> device itself for now, I think.
>
>This is exactly the same as e.g. virtio. We use Linux as a source of truth, 
>it's
>easier to share with other hypervisors this way. And UAPI and hypervisor
>ABI requirements wrt stability are mostly the same. It works.

Perfect. So as and when the header is in its final form in Linux, it can be 
part of the automated import and we'll use that version. At that point we can 
drop the one that's sitting alongside the device itself in hw/acpi/.