Re: [RFC PATCH v2 1/2] dma-buf: Add support for private interconnects

[Jason Gunthorpe <jgg@nvidia.com>]

On Mon, Sep 29, 2025 at 06:02:50PM +0200, Thomas Hellström wrote:
> > I'm not thinking about load balancing, more a 'quality of path'
> > metric.
> 
> This sounds like it's getting increasingly complex. TBH I think that at
> least all fast interconnects we have in the planning for xe either are
> fine with falling back to the current pcie-p2p / dma-buf or in worst
> case system memory. 

Yah, fallback is fine, but they will still want to prefer the better
paths if they are available. That's my point..

> The virtual interconnect we've been discussing
> would probably not be able to fall back at all unless negotiation gets
> somehow forwarded to the vm guest.

This is basically proposing to open code all the priority and QOR
matching in the drivers. I strongly think that will become
unmaintainable if things evolve in that direction..

What I'm suggesting is to have just enough infrastructure that we can
have the priority selection be in common code and implement the
simplest possible selection for right now. Run down the exports list
in order and the firt interconnect to match the importer's list is
selected.

All the more fancy stuff I've mentioned is someone else's problem down
the road to enhance the core matcher.

Jason