On Mon, Feb 09, 2026 at 14:58:56 +0000, Andrea Bolognani wrote:
> On Fri, Feb 06, 2026 at 10:27:39AM +0100, Peter Krempa wrote:
> > On Mon, Jan 12, 2026 at 15:49:16 +0100, Andrea Bolognani via Devel wrote:
> > > The current behavior is to error out, but that's not very helpful.
> > > Automatically add the necessary USB controller instead.
> >
> > IMO this creates the same situation we're in currently with the addition
> > of SCSI controllers. We add 'lsilogic' as controller if you don't have
> > any which is sub-optimal, but it's the default and the behaviour we've
> > standardized on thus we can't really change it.
> >
> > The auto-added USB controller will be likely sub-optimal either now or
> > in the future. While it seems helpful to add it it can create headaches.
> >
> > To me it seems that the users simply should add the controller manually
> > if the platform doesn't have one.
> 
> In many cases users will want to add the controller manually even for
> architectures that do provide it by default, specifically because the
> chosen model might be suboptimal. This is exactly what virt-manager
> already does, both for SCSI and USB controllers.
> 
> The fact that the behavior of automatically adding a controller when
> a device that needs it is present in the domain XML is already well
> established for SCSI and others is IMO good motivation for doing the

As I've written above I don't consider it beneficial that we add
the now-bad SCSI controller automatically.

In fact I'd personally prefer to get an (actionable) error telling me to
add a controler instead of adding the wrong one and claiming success.
It might seem like a good user experience to add one but in the end the
configuration will be worse than if the user did it manually.

> same with USB controllers, even if there are some known potential
> pitfalls with the approach.

Thus I don't think we should extend this IMO wrong approach to things.

Reply via email to