On 09/27/2018 01:51 PM, Cornelia Huck wrote:
> On Thu, 27 Sep 2018 13:29:43 +0200
> Halil Pasic <pa...@linux.ibm.com> wrote:
>
>> On 09/27/2018 12:42 AM, Alex Williamson wrote:
>>> On Tue, 25 Sep 2018 19:16:41 -0400
>>> Tony Krowiak <akrow...@linux.vnet.ibm.com> wrote:
>>>> + This is how the matrix is configured for Guest2:
>>>> +
>>>> + echo 5 > assign_adapter
>>>> + echo 0x47 > assign_domain
>>>> + echo 0xff > assign_domain
>>>> +
>>>> + This is how the matrix is configured for Guest3:
>>>> +
>>>> + echo 6 > assign_adapter
>>>> + echo 0x47 > assign_domain
>>>> + echo 0xff > assign_domain
>>>> +
>>>
>>> I'm curious why this interface didn't adopt the +/- notation invented
>>> above for consistency. Too difficult to do rollbacks with a string on
>>> entries?
>>>
>>
>> I remember that we did discuss that possibility around v9, but I can't
>> tell why did we decide to not implement it. Maybe Tony has an answer.
>
> IIRC, that was a discussion on the base ap driver interfaces rather
> than vfio-ap.
>
>>
>> Anyway, if we were to do that, we would use different attribute names
>> (e.g. just domain_mask, or something similar instead of
>> (assign|unassign)_xxx). So I think such an interface can still be added
>> on top of the existing one. Having that said having multiple interfaces
>> for the very same thing is usually not so nice IMHO.
>
> Nod to all of your points.
>
> As we do the configuration while the guest is not running anyway, the
> different interfaces probably do not make that much difference in
> practice. It should be fine to stick to the current interface for now
> and only add a new one if we really think it is significantly better.
Tony, can you maybe provide a quick on-top patch that clarifies Alex
comments regarding the documentation? (State that is is big endian,
fixup the small things etc).
I can then either fold it in or provide it as an on top patch depending
on how much has changed.
