On 27 October 2012 04:00, haibin wang <wanghaibin202@xxxxxxxxx> wrote: > There is a doubt about qemu ! > > At present, when the host kernel (kvm) didn't configure CONFIG_KVM_ARM_VGIC, > qemu load guest kernel(without CONFIG_ARCH_TIMER configuration) will be failure, > and print: > "KVM Cortex-A15 without in-kernel VGIC is not a > valid configuration. Check that: > * you have passed -machine kernel_irqchip=on to QEMU > * your host kernel was built with CONFIG_KVM_ARM_VGIC > * the device tree passed to your host kernel has a " > valid node for the GIC" > > In my opinion, loading guest kernel using GIC by qemu's simulating > must be successful . (I remeber that the host kernel of brach v5 > didn't simulate VGIC in kvm. At that > time, we loaded the guest kernel using GIC which qemu simulated). We've discussed this before. The problem is that it's not possible for emulation of the architected cp15 timer to be done in userspace [there is no way to trap the cp15 accesses so they can be forwarded to userspace]. And an in-kernel cp15 timer implementation requires an in-kernel GIC implementation. So it seemed easier to insist on this configuration (which is what everybody will be running anyway for performance reasons) rather than try to support an odd config that doesn't make much sense. As it happens, the current RFC set of QEMU patches don't insist on the in-kernel irqchip, because I wanted them to at least vaguely work (for guests that don't touch the cp15 timers) with Christoffer's base-level kernel patches that we're trying to upstream. But when the VGIC support goes upstream I'll probably reintroduce this check and error message into the QEMU side. -- PMM _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm
