On Tue, Jul 03, 2018 at 11:06:00AM +0200, Paolo Bonzini wrote: > On 03/07/2018 10:48, Robert Hoo wrote: > >> > >> However, I suggest adding it to the FeatureWord enum, since everything > >> that handles FeatureWord applies to this new kind of MSR as well. > >> Currently FeatureWord is only for CPUID leaves, but it doesn't have to > >> be like that. > >> > > I think this will be changing struct FeatureWordInfo, which is designed > > for cpuid enumerations. You must not want to do that. May I know more > > details about your thought? > > The simplest way is to put CPUIDs first and MSRs second in FeatureWord. > Then you can do > > FEAT_XSAVE_COMP_LO, /* CPUID[EAX=0xd,ECX=0].EAX */ > FEAT_XSAVE_COMP_HI, /* CPUID[EAX=0xd,ECX=0].EDX */ > + FEATURE_WORDS_NUM_CPUID, > + FEATURE_WORDS_FIRST_MSR = FEATURE_WORDS_NUM_CPUID, > + FEAT_MSR_ARCH_CAPABILITIES = FEATURE_WORDS_FIRST_MSR, > FEATURE_WORDS, > }; > > #define FEATURE_WORDS_NUM_MSRS (FEATURE_WORDS - \ > FEATURE_WORDS_FIRST_MSR) > > Then the existing loops that use FeatureWordInfo can go up to > FEATURE_WORDS_NUM_CPUID.
I assume we want to make some (or most) of the loops go up to FEATURE_WORDS (e.g. make x86_cpu_get_supported_feature_word() support MSRs too), otherwise it would be pointless to reuse the same array. I would be OK with both approaches, though. If the first version doesn't use the `features[]` array and implements this with a separate `msr_features[]` or `msr_arch_capabilities` field, it would work for me (especially if this means we can get this implemented in time for QEMU 3.0). -- Eduardo
