[libvirt] [PATCH 00/18] Enhance virpcimock and test VFIO instead of KVM

[Michal Prívozník]

On 8/16/19 11:25 PM, Daniel Henrique Barboza wrote:
> 
> 
> On 8/16/19 5:59 AM, Michal Privoznik wrote:
>> On 8/14/19 5:14 PM, Daniel Henrique Barboza wrote:
>>> I've run make check with each individual patch, and everything
>>> seems fine in my environment.
>>>
>>> For all patches:
>>>
>>> Tested-by: Daniel Henrique Barboza <danielhb413 at gmail.com>
>>>
>>>
>>> I'll see if I can drop some code reviews later on.
>>
>> That'd be great because we are lacking reviewers (just like other
>> projects) and these are very specific.
> 
> Just dropped reviews in all of them. A few comments in a couple of patches
> that you can amend before pushing if you think it's worth it (not worth
> another spin, in my opinion).
> 
> I couldn't help but wonder: can't we just remove the KVM stub (pci-assign)
> support from the code base entirely? QEMU dropped it in 2.11. The kernel
> dropped it in 4.12 (mid of 2017). Not sure if there is any existing,
> running
> guests relying on pci-assign these days that can't move to vfio-pci.
> 
> If there are still users for pci-assign, perhaps a deprecation notice is in
> order (a warning when launching the VM?). Then users can't complain
> that the support was removed and they were thrown under the bus.
> 
> 
> (... I mean, they will still complain about it, but at least we warned
> about it)
> 

That is the direction we will eventually go. I've discussed this couple
of times (and in fact I have patches ready) and it was always agreed
upon. But only yesterday I've merged a patch that fixes a bug in the way
we reset PCI devices for KVM assignment, which sparked a discussion:

  https://www.redhat.com/archives/libvir-list/2019-August/msg00671.html

But as soon as we get green light from guys I will post patches.

Michal