On 5 January 2015 at 17:34, Andre Przywara <andre.przywara@xxxxxxx> wrote: > Mmmh, fair point. I guess we should just allow KVM_MAX_VCPUS in this > case, right? Has anyone tested userland GIC emulation recently? > I will see if I find some method of testing this with QEMU. I did actually use this when I was trying to track down the bugs that turned out to be vgic-init-ordering last month. If you add -machine kernel_irqchip=off to your QEMU command line it will try to use an emulated userspace GIC. But as Christoffer says, there is a catch: there's no way to get the vcpu's generic timers to tell userspace that there's been a timer interrupt that needs to be fed into the userspace GIC. So you need to fiddle with the device tree to tell the guest the generic timers don't exist, and provide a different timer, at least in theory [though I didn't actually end up having to do that, I think it was just fluke that it worked for my testing]. -- PMM _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
