Re: [Qemu-devel] [PATCH 1/7] target/i386: handle filtered_features in a new function mark_unavailable_features

[Eduardo Habkost]

On Sat, Jul 06, 2019 at 12:07:29AM +0200, Paolo Bonzini wrote:
> On 05/07/19 23:44, Eduardo Habkost wrote:
> > On Fri, Jul 05, 2019 at 11:32:07PM +0200, Paolo Bonzini wrote:
> >> On 05/07/19 22:37, Eduardo Habkost wrote:
> >>> Filtering and reporting is separate because
> >>> x86_cpu_filter_features() is also called from a QMP command
> >>> handler that is not supposed to generate any warnings on stderr
> >>> (query-cpu-model-expansion).
> >>
> >> But that one should not set check_cpuid or enforce_cpuid, should it?
> > 
> > check_cpuid is set to true by default.
> 
> Ok, that's what I missed.
> 
> >>
> >> (I can still split the filtering and reporting if you prefer).
> > 
> > Maybe it will work if we just add a 'bool verbose' parameter to
> > x86_cpu_filter_features().
> > 
> > x86_cpu_realizefn() would call:
> >   x86_cpu_filter_features(cpu, cpu->check_cpuid);
> 
> ... "|| cpu->enforce_cpuid".
> 
> > x86_cpu_class_check_missing_features() would call:
> >   x86_cpu_filter_features(cpu, false);
> 
> Or set check_cpuid to false there after creating the object?

It would work too, but I prefer to make the side effects of
x86_cpu_filter_features() more explicit.

-- 
Eduardo