[PATCH v3 0/2] VFIO mdev aggregated resources handling
Yan Zhao
yan.y.zhao at intel.com
Wed Jul 8 03:38:46 UTC 2020
On Wed, Jul 08, 2020 at 09:54:19AM +0800, Zhenyu Wang wrote:
> On 2020.07.07 19:06:34 -0600, Alex Williamson wrote:
> > On Tue, 7 Jul 2020 23:28:39 +0000
> > "Tian, Kevin" <kevin.tian at intel.com> wrote:
> >
> > > Hi, Alex,
> > >
> > > Gentle ping... Please let us know whether this version looks good.
> >
> > I figured this is entangled with the versioning scheme. There are
> > unanswered questions about how something that assumes a device of a
> > given type is software compatible to another device of the same type
> > handles aggregation and how the type class would indicate compatibility
> > with an aggregated instance. Thanks,
> >
>
> +Yan
>
> Alex, If no concern on aggregated resources info for instance that would
> be vendor's behavior to determine what type of resources would be aggregated,
> then I'll check with Yan to see how to fulfill this during migration.
>
> Thanks
>
hi zhenyu and Alex
currently in this series, it looks that aggregated instances are created
in this way:
echo "<uuid>,aggregate=10" > create
Is that possible that we change it like that:
1. provide a separate attribute named "aggregator" under mdev type.
|- [parent physical device]
|--- Vendor-specific-attributes [optional]
|--- [mdev_supported_types]
| |--- [<type-id>]
| | |--- create
+ | | |--- aggregator
| | |--- name
| | |--- available_instances
| | |--- device_api
| | |--- description
| | |--- [devices]
normally, the aggregator is read as 0.
2. when we want to create an aggregated instance, we first echo the count
into the aggregator attribute. e.g.
echo 10 > aggregator
It will switch the mdev type to 10 x original_type. And then,
available_instances and description would be updated accordingly.
3. do the real mdev creation.
echo <uuid> > create
In this way, before any instance is created, we can use the
migration_version attribute to test if two aggregation mdevs are
migration compatible.
Thanks
Yan
More information about the intel-gvt-dev
mailing list