Re: [PATCH RFC v2 00/18] Add VFIO mediated device support and DEV-MSI support for the idxd driver

[Paolo Bonzini]

On 18/08/20 18:49, Jason Gunthorpe wrote:
> On Tue, Aug 18, 2020 at 06:27:21PM +0200, Paolo Bonzini wrote:
>> On 18/08/20 13:50, Jason Gunthorpe wrote:
>>> For instance, what about suspend/resume of containers using idxd?
>>> Wouldn't you want to have the same basic approach of controlling the
>>> wq from userspace that virtualization uses?
>>
>> The difference is that VFIO more or less standardizes the approach you
>> use for live migration.  With another interface you'd have to come up
>> with something for every driver, and add support in CRIU for every
>> driver as well.
> 
> VFIO is very unsuitable for use as some general userspace. It only 1:1
> with a single process and just can't absorb what the existing idxd
> userspace is doing.

The point of mdev is that it's not 1:1 anymore.

Paolo

> So VFIO is already not a solution for normal userspace idxd where CRIU
> becomes interesting. Not sure what you are trying to say?
> 
> My point was the opposite, if you want to enable CRIU for idxd then
> you probably need all the same stuff as for qemu/VFIO except in the
> normal idxd user API.
> 
> Jason
>