Re: [Qemu-devel] [kvm-unit-tests PATCH 00/10] arm/arm64: add gic framework

[Christoffer Dall]

On Tue, May 24, 2016 at 02:23:43PM +0200, Andrew Jones wrote:
> On Tue, May 24, 2016 at 01:58:19PM +0200, Christoffer Dall wrote:
> > On Mon, May 23, 2016 at 05:24:23PM +0200, Andrew Jones wrote:
> > > On Wed, May 18, 2016 at 11:07:14AM +0200, Christoffer Dall wrote:
> > > > Hi Drew,
> > > > 
> > > > Thanks for doing this.  I'm happy to see some tests for the GIC.
> > > > 
> > > > I've been pondering with how to write unit tests for all the MMIO
> > > > implementations.  If you have some thoughts on how that could be easily
> > > > fitted into this framework, that would probably be a good place to do it
> > > > ;)
> > > 
> > > Hi Christoffer,
> > > 
> > > Sorry for my slow response, I've been on vacation. For MMIO
> > > implementations, are you referring to the emulation done for
> > > gicv2 accesses and for gicv3 legacy accesses? And, if so, is
> > > your question how we might be able to use the same test
> > > framework for both? And, if that's so, then I think this series
> > > gets us pretty close already. If I'm completely off-base, then
> > > please give me a quick high-level description of what you'd like
> > > to be able to do.
> > > 
> > What I meant was testing all the MMIO accesses to the various
> > distributor MMIO regions.
> > 
> > For example, writing full words to all registers (some value) reading
> > back the value, correcting for RAZ/WI semantics, and testing that byte
> > accesses to those registers where that's allowed also works.
> 
> OK, understood. We can build a table that describes each distributor
> offset's allowed access types and expected read-back results for the
> "default enablement" of the gic. Then, we'd run through that table
> doing a refresh of the gic enabling before each offset test. This
> series provides everything needed for that, except the offset table.
> It should be pretty easy to add.
> 
> Now, configuring the gic differently will result in some offsets
> producing different values, so we'll eventually want to extend the
> table to check the same offsets using different gic enable functions
> as well, but that would be pretty easy to do too.
> 
> > 
> > If adding that on top of this series sounds like a good idea, someone
> > should add it to the bottom of their (presumably already long) todo
> > list, myself included.
> 
> They do sound like good tests to have. I've added it to the middle
> of my long TODO. If somebody beats me to it, I won't complain :-)
> 
Awesome, thanks!

-Christoffer