Re: [Qemu-devel] [RFC PATCH] hw/arm/virt: Added preliminary GICv3 support for kvm mode

[Pavel Fedin <p.fedin@samsung.com>]

 Hello!

> So to me it is sensible to instantiate GICV2 through legacy
> KVM_CREATE_IRQCHIP API if both KVM_CREATE_DEVICE(test mode=true) failed.

 I disagree because at this point we already know which GIC version the user wants. This
is because kvm_irqchip_create() is called after machine instance is created (and
virt_instance_init() has been called). At this point we already know all the options. At
this point i think the scenario should be:
 a) If we want GICv3 - test for KVM_CREATE_DEVICE(GICv3) and fail if we don't have one.
 b) If we want GICv2 - test for KVM_CREATE_DEVICE(GICv2). If it fails, try
KVM_CREATE_IRQCHIP.
 IMHO there is little sense to fall back from v3 to v2 or vice versa because other
important parameters (like number of CPUs) depend on it.
 Implementing this behavior costs only one more integer in MachineState structure. Is it
too large ? If you want, i can post my patches as RFC, i think now they are more or less
OK.

Kind regards,
Pavel Fedin
Expert Engineer
Samsung Electronics Research center Russia