Re: [Qemu-devel] [PATCH/RFC 2/3] s390x/ais: enable ais when migration is available

[Christian Borntraeger <borntraeger@de.ibm.com>]

On 09/25/2017 01:45 PM, Cornelia Huck wrote:
> On Mon, 25 Sep 2017 12:12:49 +0200
> Christian Borntraeger <borntraeger@de.ibm.com> wrote:
> 
>> On 09/25/2017 12:07 PM, Cornelia Huck wrote:
>>> On Fri, 22 Sep 2017 16:27:00 +0200
>>> Halil Pasic <pasic@linux.vnet.ibm.com> wrote:
>>>   
>>>> One thing I would find very helpful is what do we expect to work and not
>>>> work for which version. Kind of a matrix. For instance should vfio pci
>>>> work for versions prior 2.11. I think in the not so distant past we
>>>> changed how SIC works (so it complains when we don't have ais).  
>>>
>>> A matrix sounds like a good idea.  
>>
>> I think we do not even need a matrix, a minimum level will suffice because...
>>>
>>> I don't think we really ever had a setup that worked out of the box  
>>
>> exactly: ...it never worked until 2.10 and we do not have libvirt support yet.
>> Now with the fix 2.10 will also not work, so I think its fair to say
>>
>> PCI passthrough via VFIO will be supported for 
>> - KVM: host kernel >= 4.13
>> - TCG: TBD
>> - QEMU >= 2.11
>> - libvirt TBD
> 
> Make that zpci-per-se, no?
> 
> with KVM: host kernel >= 4.13 && QEMU >= 2.11
> with TCG: tbd, I don't think anybody has time to wire this up for 2.11
> 
> Apropos libvirt: How will it determine whether zpci should be
> supported? There are some old QEMU + KVM combinations out there that
> will have a phb (but not be usable by stock Linux guests as the feature
> bits are missing). Version fence? Check for cpu feature support?

I think for multibus or something like that Boris wanted to check for a version
anyway. So maybe 2.11 (now that 2.10 is broken regarding ais) as a minimum QEMU
level would make sense.