On Wed, Jun 17, 2020 at 4:38 AM Vitaly Kuznetsov <vkuznets@xxxxxxxxxx> wrote: > 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. So, how is MSR_IA32_PERF_CAPABILITIES different from, say, MSR_K7_HWCR, which by its very name doesn't sound like it would be supported on Intel CPUs? Why not just emulate MSR_IA32_PERF_CAPABILITIES on AMD, just as we emulate MSR_K7_HWCR on Intel? My concern is that we don't seem to have a standard here. Each individual MSR is handled ad hoc, which adds unnecessary complexity.
