On 4 December 2020 18:26:54 GMT, Alexander Graf <graf@xxxxxxxxxx> wrote: >Why does any of this have to live in kernel space? I would expect user >space to own the MSR and hcall number space and push it down to KVM. >That means it can also provide the shims it needs. Because the kernel is already trapping the VMCALL and handling all the KVM hypercalls, because it's much more efficient to do so. This is just a way to allow userspace to see those exits and handle the hypercalls *instead* of the kernel. I suppose you can make the case that none of the KVM hypercalls should be handled in-kernel and it could all be done by trapping to userspace. But that ship sailed long ago, even if you were right and the additional latency of handling the KVM hypercalls was indeed acceptable. -- Sent from my Android device with K-9 Mail. Please excuse my brevity.
