NVIDIA GPU passthrough into Linux HVMs for CUDA applications

The instructions appear to have a problem with them, in that you are using the kernel from dom0:

but then akmod will end up compiling against the headers of the kernel version installed in the VM, and complaining.

Did using the dom0 kernel, but compiling against the headers of the kernel version installed in the fedora VM actually work for you, or did you forget to document a step?