[PATCH] arm: introduce a DTS for Xen unprivileged virtual machines

[Ian Campbell]

On Thu, 2012-09-20 at 12:56 +0100, Stefano Stabellini wrote:
> On Thu, 20 Sep 2012, Ian Campbell wrote:
> > On Wed, 2012-09-19 at 18:44 +0100, Stefano Stabellini wrote:
> > > +/include/ "skeleton.dtsi"
> > > +
> > > +/ {
> > > +	model = "XENVM-4.2";
> > 
> > Why the shouty caps?
> 
> It looks like that model names are always capital, at least in the
> vexpress family.
> 
> 
> > Did you mean 4.3 here and throughout?
> 
> Nope, after all this is the fruit of the work we did on Xen 4.2, mostly
> already upstream. By the time of the 4.3 release we might have a
> different dts.
> 
> 
> > > +	compatible = "xen,xenvm-4.2", "arm,vexpress";
> > 
> > Is this second compatible thing actually true? We don't actually emulate
> > much (anything?) of what would be on a real vexpress motherboard.
> > 
> > "arm,vexpress" is used only in v2m.c and I don't think we want the
> > majority of that -- we don't provide any of the peripherals which it
> > registers.
> > 
> > I think the only things we might want out of that lot are the arch timer
> > and perhaps the uart0 (as a debug port).
> > 
> > I suspect we should have our own xen machine .c.
> > 
> > [...]
> 
> It is true that we are "arm,vexpress" compatible at the moment.

But we aren't, we don't emulate 90%+ of the actual hardware which
vexpress compatibility would actually imply.

Look in arch/arm/mach-vexpress/v2m.c, which is the only thing keyed off
this compat value -- it's full of stuff which we don't (and aren't going
to) implement.

> Also we need to be unless we want to introduce our own arch/arm/mach-xen
> that I think is overkill.

Probably.

> Versatile Express is flexible enough to be a good base for our own
> virtual machine platform, especially if the maintainers keep an eye on
> getting everything through DT and not expecting devices just to be there
> ;-)

Perhaps what we want is a stricter subset of the stuff in mach-vexpress
then. But if so then this should be expressed both in the DT and in the
code, not just papered over by declaring things compatible when they are
not.

> > > +	gic: interrupt-controller at 2c001000 {
> > > +		compatible = "arm,cortex-a9-gic";
> > 
> > Don't we mean "arm,cortex-a15-gic" here? That's what we actually
> > provide. I'm not sure how the a9 and a15 differ.
> 
> The GIC that comes with vexpress is a9 compatible.

The GIC which Xen emulates is the one which matters here though, and
that is an a15.

Ian.