On 6/23/23 15:12, Boris Fiuczynski wrote:
> On 6/21/23 6:54 PM, Jonathon Jongsma wrote:
>> On 6/13/23 10:42 AM, Boris Fiuczynski wrote:
>>> Update capabilities for QEMU 8.1 on s390x, add a new capability
>>> async-teardown and make use of it when running on s390x hosts to improve
>>> memory reclaiming.
>>
>> Is this really something that should be enabled unconditionally on all
>> s390x guests, or should it be configured with some domain xml? If
>> there's ever a case where an s390x domain would want this disabled, I
>> think it would have to be configurable. Also, if there is any
>> situation where a domain on a different architecture might want to
>> enable this, that would also require some kind of configurability. At
>> minimum it seems to me that the commit log should have a lot more
>> justification for why this approach is justified.
>>
>> Jonathon
> 
> Jonathon,
> thanks for your feedback.
> 
> I am unsure where to located such a configuration option in the gust
> domain XML.
> 
> A few thoughts:
> 1) introduce a new emulator-options element in devices and a run-with
> child element with a parameter async-teardown, e.g.
> ...
> <devices>
>   <emulator>/usr/lib/bin/qemu</emulator>
>   <emulator-options>
>     <run-with async-teardown='on'/>
>   </emulator-options>
> </devices>
> ...
> 
> 2) introduce a new run-with element with a parameter async-teardown in
> domain, e.g.
> 
> <domain>
>   ...
>   <run-with async-teardown='on'/>
>   ...
> </domain>
> 
> Any ideas and suggestions are welcome.

We usually use domain <features/>, e.g.:

<domain>
  <name>someName</name>
  <uuid/>
  ...
  <features>
    <acpi/>
    <tcg>
       <tb-cache unit='KiB'>102400</tb-cache>
    </tcg>
  </features>
  ...
  <devices/>
</domain>

This looks like a fit place if we want to expose this in the domain XML.
And from the qemu-options.hx file it doesn't look s390x specific, which
means, we don't need to make the new element arch specific. IOW we can have:

  <features>
    <async-teardown enabled='yes'/>
  </features>

Michal

Reply via email to