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

[Marc Zyngier]

On 20/09/12 13:15, Stefano Stabellini wrote:
> On Thu, 20 Sep 2012, Pawel Moll wrote:
>> On Thu, 2012-09-20 at 12:39 +0100, Stefano Stabellini wrote:
>>> There are no peripherals apart from the ones that are already described
>>> here (timer, gic). All the peripherals that the guest sees are virtual
>>> devices that show up on xenbus (a virtual bus). In order to initialize
>>> xenbus, the guest only needs the hypervisor node.  So I'll remove the
>>> ranges and interrupt-map.
>>
>> So all the peripherals are actually discoverable - awesome!
>>
>> But the fact is that the only "vexpressness" of this tree is
>> memory at 80000000 and gic at 2c001000. The rest (not much of it ;-) is a
>> "generic A15 platform". 
>>
>> I understand that you had to use some "compatible" value to get
>> initialization code and I'm flattered ;-) by your choice of
>> "arm,vexpress", but maybe - as suggested by others - you would be better
>> off with generating this stuff in runtime, by whatever tool is used to
>> instantiate the guest?
> 
> Yes, it is certainly going to be generated at run time. Does it make
> sense to keep an example under arch/arm/boot/dts? I think so. There must
> be other platforms that actually pass the device tree to the kernel from
> the firmware and still have a dts in the kernel tree, right?
> If actually there are none, it is probably best to forget about this
> patch :)
> 
> Regarding the choice of "arm,vexpress", there was a discussion at kernel
> summit about vexpress being the right choice as a base platform for
> virtual machines on ARM, even though in the Xen case it means having a
> vexpress with no peripherals.

I think the important thing here is that the memory map is RS1. As this
is a (very limited) subset of a vexpress A15, it seems to make some sense.

I still think it is useful at have an example of such a platform in the
tree, if only as a very minimal example.

	M.
-- 
Jazz is not dead. It just smells funny...