On Wed, Jan 10, 2024 at 12:57:41PM +0100, Peter Krempa wrote:
On Mon, Jan 08, 2024 at 20:26:01 +0100, Andrea Bolognani wrote:
For machines that don't expose useful information through sysfs, the dummy ID 0 is used.
I'm not a fan, as you won't be able to use the capabilities XML to figure out whether it's supported or not, but it's the same case with 'die_id' which we approach exactly the same.
I think the idea is that any management application that is going to look into the capabilities information to try and optimize guest vCPU placement will see that there's a single die/cluster and thus conclude that there are no opportunities for optimization there. So in practice whether die/cluster support is present doesn't really matter. But yeah, I've just followed the same approach established for dies and I wouldn't want to deviate from it, as doing so would probably just create confusion.
Also there's no docs about those fields either in the capability XML docs.
Based on the fact that this follows what we allowed for 'die_id' I can't really reasonably request the docs, but I'd really welcome if we had some.
I've taken a stab at it in v2. -- Andrea Bolognani / Red Hat / Virtualization