Re: [Qemu-devel] No more chameleon devices (was: [PATCH] virtio: Provide version-specific variants of virtio PCI devices)

["Michael S. Tsirkin" <mst@redhat.com>]

On Wed, Oct 17, 2018 at 12:56:37PM -0300, Eduardo Habkost wrote:
> (CCing Marcel, in case he has extra details on the complex
> Conventional/Express bus/device plugging rules)
> 
> On Wed, Oct 17, 2018 at 07:57:39AM +0200, Markus Armbruster wrote:
> > Laine Stump <laine@redhat.com> writes:
> > 
> > [...]
> > > In the end, having a device that changed PCI ID depending on what kind
> > > of slot it was plugged into was an idea "too clever for its own good",
> > > should be avoided when new devices are added in the future, and we
> > > should at least provide an alternative that doesn't do that for existing
> > > devices.
> > 
> > That means for each chameleon PCI/PCIe device:
> > 
> > * create a pair of devices that can only go into one kind of slot
> > 
> > * deprecate the chameleon
> > 
> > Yes, please!  Volunteers?
> > 
> > Do we have similar chameleons outside PCI?
> 
> I'm worried that we could be trying to address multiple issues at
> the same time, and I'm not sure yet if we should address all of
> them in one take.
> 
> Right now we need to differentiate non-transitional and
> transitional virtio devices, for a few reasons:
> * They have different PCI IDs;
> * Legacy drivers don't work with non-transitional devices;
> * Transitional virtio devices can be plugged to Conventional PCI
>   buses; non-transitional ones can't.

No that last point isn't true at all.

> This patch addresses that problem.
>
> You seem to be talking about a different issue:
> * Some devices (including transitional virtio) can be plugged on
>   both Conventional PCI and PCI Express buses (I will call those
>   devices "hybrid PCI devices").
> 
> The former is a practical problem: management software needs to
> be able to ask for a transitional virtio device, depending o the
> guest OS being run.
> 
> Addressing the latter seems more complex (it would affect other
> devices, not just virtio), and I don't see which practical
> problems it would solve.
> 
> I see some problems it wouldn't solve, though: the system
> wouldn't be able to represent the fact that transitional virtio
> devices can still work on PCI Express buses, as long as they
> support PIO bars; or that Conventional PCI devices can be plugged
> to PCI Express root buses.
> 
> I don't see problems caused by hybrid conventional/express PCI
> devices.  The original problem with virtio devices was just not
> being hybrid, it was lying about being hybrid: non-transitional
> virtio devices are hybrid, but transitional virtio devices
> aren't.
> 
> I wouldn't be against abolishing hybrid PCI devices completely if
> somebody volunteers to do the work.  I just don't see which
> problems this would solve.
> 
> -- 
> Eduardo