On 28/06/19 16:06, Michael S. Tsirkin wrote: >> + assert(kvm_irqchip_in_kernel()); > Hmm - irqchip in kernel actually increases the attack surface, > does it not? Or at least, the severity of the attacks.
Yeah, we should at least support split irqchip. But, irqchip completely in userspace is slow when it is not broken, and it does not support APICv. So it's not really feasible. Paolo
