[Qemu-devel] [kvm-unit-tests PATCH v7 00/11] QEMU MTTCG Test cases

[Andrew Jones]

On Mon, Nov 28, 2016 at 03:04:45PM +0100, Andrew Jones wrote:
> On Mon, Nov 28, 2016 at 01:30:54PM +0000, Peter Maydell wrote:
> > On 28 November 2016 at 11:58, Andrew Jones <drjones at redhat.com> wrote:
> > > On Mon, Nov 28, 2016 at 11:14:48AM +0000, Peter Maydell wrote:
> > >> On 28 November 2016 at 11:12, Alex Bennée <alex.bennee at linaro.org> wrote:
> > >> >
> > >> > Andrew Jones <drjones at redhat.com> writes:
> > >> >> I've skimmed over everything looking at it from a framwork/sytle
> > >> >> perspective. I didn't dig in trying to understand the tests though.
> > >> >> One general comment, I see many tests introduce MAX_CPUS 8. Why do
> > >> >> that? Why not allow all cpus by using NR_CPUS for the array sizes?
> > >> >
> > >> > Yeah - I can fix those. I wonder what the maximum is with GIC V3?
> > >>
> > >> So large that you don't want to hardcode it as an array size...
> > >
> > > 255 with the gic series, not yet merged.
> > 
> > I was talking about the architectural GICv3 limit, which is larger
> > than that by many orders of magnitude. For QEMU it looks like
> > MAX_CPUMASK_BITS is now 288 rather than 255.
> 
> Ah, yeah. So far we haven't considered testing limits beyond what
> KVM supports, VGIC_V3_MAX_CPUS=255. However with TCG, and some
> patience, we could attempt to test bigger limits. In that case,
> though, we'll want to recompile kvm-unit-tests with a larger NR_CPUS
> and run a specific unit test.
> 
> mach-virt still has 255 as well, mc->max_cpus = 255, so we'd have
> to bump that too if we want to experiment.

Er... actually mach-virt is 123, as we only allocate 123 redistributors.

drew