Re: Surely cpumodel isn't cumulative?

[Stephen Borrill <netbsd%precedence.co.uk@localhost>]

On Mon, 30 Nov 2009, Christoph Egger wrote:
> > On Mon, 30 Nov 2009, Stephen Borrill wrote:
> > > > From i386/coretemp.c:
> > >
> > >        cpumodel = CPUID2MODEL(ci->ci_signature);
> > >        /* extended model */
> > >        cpumodel += CPUID2EXTMODEL(ci->ci_signature);
> > >
> > > But EXTMODEL is extra detail within a given model, they
> > > can't be added together.
> > >
> > > i.e. model 0xe with ext model 0x1 is not the same as 0xf.
> > >
> > > Any unforeseen issues with just removing the second cpumodel
> > > line (or altering it to cpuextmodel)?
> >
> > This patch works for me on Xeon X3430:
> >
> > --- coretemp.c  25 Mar 2009 22:53:51 -0000      1.12
> > +++ coretemp.c  30 Nov 2009 15:28:40 -0000
> > @@ -67,7 +67,7 @@
> >          struct coretemp_softc *sc;
> >          uint32_t regs[4];
> >          uint64_t msr;
> > -       int cpumodel, cpumask;
> > +       int cpumodel, cpuextmodel, cpumask;
> >
> >          /*
> >           * CPUID 0x06 returns 1 if the processor has on-die thermal
> > @@ -85,7 +85,7 @@
> >              (int)device_unit(ci->ci_dev));
> >          cpumodel = CPUID2MODEL(ci->ci_signature);
> >          /* extended model */
> > -       cpumodel += CPUID2EXTMODEL(ci->ci_signature);
> > +       cpuextmodel = CPUID2EXTMODEL(ci->ci_signature);
> >          cpumask = ci->ci_signature & 15;
>
> On AMD you would write here:
>
>   if (cpumodel == 0xf)
>       cpumodel += cpuextmodel;
>
> and the hunk below would work unchanged.
> Check the Intel docs if you can do that on Intel, too.

No, you can't:

See page 16:
http://www.intel.com/Assets/en_US/PDF/specupdate/322373.pdf

--
Stephen