Jim Mattson <jmattson@xxxxxxxxxx> writes: > On Tue, Jun 16, 2020 at 9:45 AM Vitaly Kuznetsov <vkuznets@xxxxxxxxxx> wrote: >> >> Jim Mattson <jmattson@xxxxxxxxxx> writes: >> >> > On Tue, Jun 16, 2020 at 9:14 AM Vitaly Kuznetsov <vkuznets@xxxxxxxxxx> wrote: >> >> >> >> state_test/smm_test selftests are failing on AMD with: >> >> "Unexpected result from KVM_GET_MSRS, r: 51 (failed MSR was 0x345)" >> >> >> >> MSR_IA32_PERF_CAPABILITIES is an emulated MSR indeed but only on Intel, >> >> make svm_has_emulated_msr() skip it so it is not returned by >> >> KVM_GET_MSR_INDEX_LIST. >> > >> > Do we need to support this MSR under SVM for cross-vendor migration? >> > Or, have we given up on that? >> >> To be honest I'm not sure about the status of cross-vendor migration in >> general and PMU implications in particular, hope Paolo/Sean can shed >> some light. In this particular case my shallow understanding is that >> MSR_IA32_PERF_CAPABILITIES has only one known feature bit which unlocks >> an MSR range with additional counters. If the feature bit is not set >> this, I guess, can easily be migrated (basically, let's allow writing >> '0' there on AMD and return '0' on read). But what if the feature was >> enabled? We'll have to support the new MSR range and do something with >> it after migration (run intel_pmu in fully emulated mode?). >> >> Anyway, the immediate issue I'm trying to fix here is: whatever is >> returned by KVM_GET_MSR_INDEX_LIST can be successfully queried with >> KVM_GET_MSRS as some userspaces count on that. > > That's a nice property. Is it documented somewhere? > Hm, good question. Documentation/virt/kvm/api.rst says: "KVM_GET_MSR_INDEX_LIST returns the guest msrs that are supported. The list varies by kvm version and host processor, but does not change otherwise. [...] KVM_GET_MSR_FEATURE_INDEX_LIST returns the list of MSRs that can be passed to the KVM_GET_MSRS system ioctl. This lets userspace probe host capabilities and processor features that are exposed via MSRs (e.g., VMX capabilities)." Side note: MSR_IA32_PERF_CAPABILITIES can be returned by both KVM_GET_MSR_INDEX_LIST and KVM_GET_MSR_FEATURE_INDEX_LIST as we have it both as an emulated MSR filtered by kvm_x86_ops.has_emulated_msr() and a feature msr filtered by kvm_x86_ops.get_msr_feature(). But the later is a whitelist so MSR_IA32_PERF_CAPABILITIES won't appear on AMD and the promise "can be passed to the KVM_GET_MSRS" is kept. For KVM_GET_MSR_INDEX_LIST, the promise is "guest msrs that are supported" and I'm not exactly sure what this means. Personally, I see no point in returning MSRs which can't be read with KVM_GET_MSRS (as this also means the guest can't read them) and KVM selftests seem to rely on that (vcpu_save_state()) but this is not a documented feature. > Reviewed-by: Jim Mattson <jmattson@xxxxxxxxxx> > Thanks! -- Vitaly
