Re: [RFC proxmox-perl-rs/qemu-server/qemu-server-rs v2 0/9] pci-handling rewrite (part 1/2/3)

[Dominik Csapak <d.csapak@proxmox.com>]

On 10/29/26 4:36 PM, Fiona Ebner wrote:
> Am 15.10.26 um 10:11 AM schrieb Dominik Csapak:
>> ## Phase 4 - Host and NUMA layouts (the next part)
>>
>> With all the pieces in place, we can then easily extend the new layout by PCI
>> switches that are needed for assigning to NUMA nodes and replicating the host
>> topology for passed through devices.
> Can you describe the planned VM configuration settings for this? How do
> they fit alongside/relate to the "pci-layout" option you introduce in
> patch 9/9? It would be nice to know before the "pci-layout" option is
> set in stone.


i would have imagined something like:

machine: q35,pci-layout=grouped,pci-topology=<numa|host>

i'd use new device types in the grouped layout to represent the pci 
expanders necessary for numa nodes and switches etc, to replicate
either the (reduced) host topology of the devices
or simply to add them to the relevant numa node

alternatively we could add new pci-layouts for these two
types of topologies and having them strictly separate

depends a bit on what is more practical:

* another config on the perl side but having a single layout
* having more layouts but only a single config option

i'd send a new version of this and the topology things after
that next week, i think then it's easier to see how it
would look like