On 18.04.19 13:31, Christian Borntraeger wrote: > While we have removed csske and bpb from the default model, the very > common case of "host-model" in libvirt (expanded to a base model > + features) would still contain bpb and csske. This can prevent > migration to a future machine for a host-model machine. Let us fence > bpb and csske when we run on a generation 15. If necessary the > user can still use -cpu 8561,csske=on,bpb=on to force enable these > features. > > Signed-off-by: Christian Borntraeger <[email protected]> > --- > target/s390x/cpu_models.c | 11 +++++++++++ > 1 file changed, 11 insertions(+) > > diff --git a/target/s390x/cpu_models.c b/target/s390x/cpu_models.c > index e727519686..03bdca1fb8 100644 > --- a/target/s390x/cpu_models.c > +++ b/target/s390x/cpu_models.c > @@ -629,6 +629,17 @@ CpuModelExpansionInfo > *qmp_query_cpu_model_expansion(CpuModelExpansionType type, > return NULL; > } > > + /* > + * Do not claim CSSKE and BPB if the host model expands to > + * generation 15 or newer > + */ > + if (s390_model.def->gen >= 15) { > + clear_bit(S390_FEAT_CONDITIONAL_SSKE, > + (unsigned long *) &s390_model.features); > + clear_bit(S390_FEAT_BPB, > + (unsigned long *) &s390_model.features); > + } > + > if (type == CPU_MODEL_EXPANSION_TYPE_STATIC) { > delta_changes = true; > } else if (type != CPU_MODEL_EXPANSION_TYPE_FULL) { >
The bad thing about it is, it hinders migration *to* this hw generation. Will have to think about this some more. As alternative, simply ignore S390_FEAT_CONDITIONAL_SSKE and S390_FEAT_BPB when baselining/comparing. -- Thanks, David / dhildenb
