Sebastien Boeuf

> > The interaction with the underlying hypervisor is only one piece of the problem. There's a lot of work to ensure you can abstract the system calls between Linux...

@rbradford what can we do about this in the short term? Seems like we're gonna have to disable `test_vfio` for a while :-/ Maybe we can rely on our baremetal...

Yes seems like it, but this is going to take quite some time given the fix isn't even submitted upstream :-/

Thanks for the update on this.

The question now is to identify if this patch introduces some unwanted behavior or if the MSI-X implementation of Cloud Hypervisor lacks something.

Do you have some updates @rbradford ?

> @rbradford @sboeuf Based on VFIO USER spec, besides dirty page tracking over the guest RAM (for potential DMA operations from the device/server side), we will also need to migrate...

Before we move on with the implementation of the migration support for `vfio-user`, let's wait for a few weeks. This will allow the new VFIO migration API to land upstream...

@likebreath thanks for summarizing this. I've commented directly on your drafting commit since it was more convenient. Overall looks good, but you must update your comments about the way we're...

Hi @tmakatos, I'm curious what's the status on the adoption of VFIO migration API v2 in vfio-user? I'm just trying to understand how far you are from merging the code...