Re: [Qemu-devel] [PATCH v6 3/3] target-i386: Return runnability information on query-cpu-definitions

[Igor Mammedov <imammedo@redhat.com>]

On Mon, 10 Oct 2016 14:01:10 -0300
Eduardo Habkost <ehabkost@redhat.com> wrote:

> On Mon, Oct 10, 2016 at 02:27:49PM +0200, Igor Mammedov wrote:
> > On Fri,  7 Oct 2016 17:29:02 -0300
> > Eduardo Habkost <ehabkost@redhat.com> wrote:
> >   
> > > Fill the "unavailable-features" field on the x86 implementation
> > > of query-cpu-definitions.
> > > 
> > > Cc: Jiri Denemark <jdenemar@redhat.com>
> > > Cc: libvir-list@redhat.com
> > > Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
> > > ---
> > > Changes v5 -> v6:
> > > * Call x86_cpu_filter_features(), now that x86_cpu_load_features()
> > >   won't run it automatically
> > > 
> > > Changes v4 -> v5:
> > > * (none)
> > > 
> > > Changes v3 -> v4:
> > > * Handle missing XSAVE components cleanly, but looking up
> > >   the original feature that required it
> > > * Use x86_cpu_load_features() function
> > > 
> > > Changes v2 -> v3:
> > > * Create a x86_cpu_feature_name() function, to
> > >   isolate the code that returns the property name
> > > 
> > > Changes v1 -> v2:
> > > * Updated to the new schema: no @runnable field, and
> > >   always report @unavailable-features as present
> > > ---
> > >  target-i386/cpu.c | 76 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
> > >  1 file changed, 76 insertions(+)
> > > 
> > > diff --git a/target-i386/cpu.c b/target-i386/cpu.c
> > > index 23cc19b..63330ce 100644
> > > --- a/target-i386/cpu.c
> > > +++ b/target-i386/cpu.c
> > > @@ -1945,6 +1945,27 @@ static inline void feat2prop(char *s)
> > >      }
> > >  }
> > >  
> > > +/* Return the feature property name for a feature flag bit */
> > > +static const char *x86_cpu_feature_name(FeatureWord w, int bitnr)
> > > +{
> > > +    /* XSAVE components are automatically enabled by other features,
> > > +     * so return the original feature name instead
> > > +     */
> > > +    if (w == FEAT_XSAVE_COMP_LO || w == FEAT_XSAVE_COMP_HI) {
> > > +        int comp = (w == FEAT_XSAVE_COMP_HI) ? bitnr + 32 : bitnr;
> > > +
> > > +        if (comp < ARRAY_SIZE(x86_ext_save_areas) &&
> > > +            x86_ext_save_areas[comp].bits) {
> > > +            w = x86_ext_save_areas[comp].feature;
> > > +            bitnr = ctz32(x86_ext_save_areas[comp].bits);
> > > +        }
> > > +    }
> > > +
> > > +    assert(bitnr < 32);
> > > +    assert(w < FEATURE_WORDS);
> > > +    return feature_word_info[w].feat_names[bitnr];
> > > +}
> > > +
> > >  /* Compatibily hack to maintain legacy +-feat semantic,
> > >   * where +-feat overwrites any feature set by
> > >   * feat=on|feat even if the later is parsed after +-feat
> > > @@ -2030,6 +2051,59 @@ static void x86_cpu_parse_featurestr(const char *typename, char *features,
> > >      }
> > >  }
> > >  
> > > +static void x86_cpu_load_features(X86CPU *cpu, Error **errp);
> > > +static int x86_cpu_filter_features(X86CPU *cpu);
> > > +
> > > +/* Check for missing features that may prevent the CPU class from
> > > + * running using the current machine and accelerator.
> > > + */
> > > +static void x86_cpu_class_check_missing_features(X86CPUClass *xcc,
> > > +                                                 strList **missing_feats)
> > > +{
> > > +    X86CPU *xc;
> > > +    FeatureWord w;
> > > +    Error *err = NULL;
> > > +    strList **next = missing_feats;
> > > +
> > > +    if (xcc->kvm_required && !kvm_enabled()) {
> > > +        strList *new = g_new0(strList, 1);
> > > +        new->value = g_strdup("kvm");;
> > > +        *missing_feats = new;
> > > +        return;
> > > +    }
> > > +
> > > +    xc = X86_CPU(object_new(object_class_get_name(OBJECT_CLASS(xcc))));
> > > +
> > > +    x86_cpu_load_features(xc, &err);
> > > +    if (err) {
> > > +        /* Errors at x86_cpu_load_features should never happen,
> > > +         * but in case it does, just report the model as not
> > > +         * runnable at all using the "type" property.
> > > +         */
> > > +        strList *new = g_new0(strList, 1);
> > > +        new->value = g_strdup("type");
> > > +        *next = new;
> > > +        next = &new->next;
> > > +    }
> > > +
> > > +    x86_cpu_filter_features(xc);
> > > +
> > > +    for (w = 0; w < FEATURE_WORDS; w++) {
> > > +        uint32_t filtered = xc->filtered_features[w];
> > > +        int i;
> > > +        for (i = 0; i < 32; i++) {
> > > +            if (filtered & (1UL << i)) {
> > > +                strList *new = g_new0(strList, 1);
> > > +                new->value = g_strdup(x86_cpu_feature_name(w, i));
> > > +                *next = new;
> > > +                next = &new->next;
> > > +            }
> > > +        }
> > > +    }  
> > Shouldn't you add 
> >    if (IS_AMD_CPU(env)) { 
> > fixup here, that realize does right after calling x86_cpu_filter_features()  
> 
> What would it be useful for? The IS_AMD_CPU fixup runs after
> x86_cpu_filter_features() (so it doesn't affect filtered_features
> at all), and filtered_features is the only field used as input to
> build missing_feats.
For completeness of features returned by query-cpu-definitions, I'd guess.
So that returned cpu definitions would match actually created cpus.