Re: [PATCH v2 1/4] conf: Introduce dynamicMemslots attribute for virtio-mem
On Wed, Jan 10, 2024 at 12:36:58 +0100, Michal Prívozník wrote:
On 1/8/24 10:00, David Hildenbrand wrote:
> On 08.01.24 09:38, Peter Krempa wrote:
>> On Fri, Jan 05, 2024 at 18:30:56 +0100, David Hildenbrand wrote:
>>> On 05.01.24 14:58, Peter Krempa wrote:
>>>> On Fri, Jan 05, 2024 at 14:34:53 +0100, David Hildenbrand wrote:
>>>>> On 05.01.24 13:40, Peter Krempa wrote:
>>>>>> On Fri, Jan 05, 2024 at 13:33:32 +0100, Michal Privoznik wrote:
>>
>> [...]
>>
So what's the resolution here? IIUC, QEMU will enable this by default
and keep it off for older machine types. If that's the case, we don't
need these after all. Or am I misunderstanding something?
So, at this point automatically enabling that feature is not possible as
it makes the VM incompatible with 'vhost-user' devices, including ones
hotplugged in the future.
Based on the fact that it's about 'vhost-user' qemu won't be ever 100%
sure that the possibly hotplugged device is modern enough to be
compatible, as it depends on other software such as virtiofsd.
As of such we will most likely need the flag as a workaround for when
qemu decides to enable it, in case users want to use older software.
Regarding this patchset, the documentation needs to be improved to
clearly state the specific details about what users are supposed to,
including outlining the workaround it provides:
- if users don't want vhost-user devices (now, or hotplug) -> enable it
- if users have modern vhost-user devices -> enable it
- for now the default value is: disabled, but qemu might want to
auto-enable it in the future
- if there's a problem in the future once it's auto-enabled, disable it
explicitly if you have problems with vhost-user (e.g. virtiofs)
devices.
With the clear guidance on when to set it the flag is acceptable as it
sometimes requires policy decision (such as user wants to hotplug
virtiofs, which is not yet fixed) which we can't make automatically