> From: Qemu-arm [mailto:qemu-arm-bounces+salil.mehta=huawei....@nongnu.org] > On Behalf Of Andrew Jones > Sent: Wednesday, June 3, 2020 3:37 PM > To: Salil Mehta <salil.me...@huawei.com> > Cc: Peter Maydell <peter.mayd...@linaro.org>; m...@redhat.com; > qemu-devel@nongnu.org; eric.au...@redhat.com; qemu-...@nongnu.org; Igor > Mammedov <imamm...@redhat.com> > Subject: Re: [Question] Regarding PMU initialization within the QEMU for ARM > VCPUs > > On Wed, Jun 03, 2020 at 01:48:10PM +0000, Salil Mehta wrote: > > (Maybe I should float the ARM VCPU Hotplug patches and let this > > discussion be held over there?) > > > > Yes, I think that would be best. Keep in mind that the 'pmu' CPU property > is just one CPU property that we require all CPUs to have, if any have it. > 'aarch64' and 'sve' are two other examples. And, likely any CPU feature > that comes down the line that we want to use with KVM will fit that > pattern. I think the hotplug patch series will need to handle those > features in some way other than to push them all into machine properties.
Sure, I do have realization about that. I have for now used a per virt machine flag for PMU to do a workaround of the problem I discussed. Maybe once I float the patches further comments on the ways to improve the design would be very helpful. So for now I will keep the flag and use this approach and invite everyone for open discussion about this and we could evolve the design as we discuss there. Many thanks Salil.