Re: [Qemu-devel] [PATCH v3 3/4] hw/arm: versal: Add a model of Xilinx Versal SoC

[Edgar E. Iglesias]

On Sun, Oct 21, 2018 at 10:25:28PM +0100, Peter Maydell wrote:
> On 21 October 2018 at 20:24, Edgar E. Iglesias
> <address@hidden> wrote:
> > On Fri, Oct 19, 2018 at 03:18:24PM +0100, Peter Maydell wrote:
> >> On 17 October 2018 at 22:39, Edgar E. Iglesias <address@hidden> wrote:
> >> > From: "Edgar E. Iglesias" <address@hidden>
> >> >
> >> > Add a model of Xilinx Versal SoC.
> >> >
> >> > Signed-off-by: Edgar E. Iglesias <address@hidden>
> >> > ---
> >> >  default-configs/aarch64-softmmu.mak |   1 +
> >> >  hw/arm/Makefile.objs                |   1 +
> >> >  hw/arm/xlnx-versal.c                | 324 ++++++++++++++++++++++++++++
> >> >  include/hw/arm/xlnx-versal.h        | 122 +++++++++++
> >> >  4 files changed, 448 insertions(+)
> >> >  create mode 100644 hw/arm/xlnx-versal.c
> >> >  create mode 100644 include/hw/arm/xlnx-versal.h
> >>
> >> > +    if (!kvm_irqchip_in_kernel()) {
> >> > +        qdev_prop_set_bit(gicdev, "has-security-extensions", true);
> >> > +    }
> >>
> >> Do you really support KVM for this board/SoC ?
> >>
> >
> >
> > I haven't tried yet, so probably not, but KVM is something we'd like to 
> > support further down the road...
> > If you prefer, we can remove this kvm specific check for now though.
> 
> I think there's other things you need to do to support KVM
> (for instance, you need to disable EL2 and EL3 on all the CPUs,
> and you need to either handle a GICv2 or error out properly
> if the host system doesn't have a GICv3), so maybe it would
> be better to add support properly later. This isn't a subtle
> check we'll forget to add in later either -- if you set
> has-security-extensions on a KVM GICv3 then the device will
> fail its 'realize' method with a suitable error.

Thanks, yes, we can do that later. I'll remove this check for now.

> 
> ...which leads me to notice that here:
> 
> +    object_property_set_bool(OBJECT(&s->fpd.apu.gic), true, "realized", 
> errp);
> 
> we capture the possible error from realize in errp, but we
> don't actually check whether it failed, so the rest of the
> function will plough ahead and try to wire up IRQs and
> MemoryRegions that won't have been created.

Thanks, I'll fix that in the next version!

Cheers,
Edgar