On 23/01/20 16:27, milanpa@xxxxxxxxxx wrote: >> >> >> Paolo >> > I agree, extending the API with GET_AVAILABLE_ONE_REGS (and possibly a > bitmask argument to narrow down which type of registers userspace is > interested in) is a clean solution. We won't require userspace to rely > on constants in compile time if it doesn't need to. > > Only concern is that now we need to have some kind of datastructure for > keeping the mappings between all available ONE_REG IDs and their > strings/descriptions. Additionally enforcing that newly added ONE_REGs > always get added to that mapping, is also necessary. For now just do the implementation for VM ONE_REGs. We'll worry about the existing VCPU registers later. Paolo
