On 18/05/20 06:52, Xiaoyao Li wrote: > On 5/6/2020 5:44 PM, Paolo Bonzini wrote: >> Using CPUID data can be useful for the processor compatibility >> check, but that's it. Using it to compute guest-reserved bits >> can have both false positives (such as LA57 and UMIP which we >> are already handling) and false negatives: > >> in particular, with >> this patch we don't allow anymore a KVM guest to set CR4.PKE >> when CR4.PKE is clear on the host. > > A common question about whether a feature can be exposed to guest: > > Given a feature, there is a CPUID bit to enumerate it, and a CR4 bit to > turn it on/off. Whether the feature can be exposed to guest only depends > on host CR4 setting? I.e., if CPUID bit is not cleared in cpu_data in > host but host kernel doesn't set the corresponding CR4 bit to turn it > on, we cannot expose the feature to guest. right? It depends. The most obvious case is that the host kernel doesn't use CR4.PSE but we even use 4MB pages to emulate paging disabled mode when the processor doesn't support unrestricted guests. Basically, the question is whether we are able to save/restore any processor state attached to the CR4 bit on vmexit/vmentry. In this case there is no PKRU field in the VMCS and the RDPKRU/WRPKRU instructions require CR4.PKE=1; therefore, we cannot let the guest enable CR4.PKE unless it's also enabled on the host. Paolo
