Re: Detect on-die temp sensor for Atom E6xx on amd64

[Matt Dainty]

* Christian Weisgerber <na...@mips.inka.de> [2013-01-24 13:03:43]:
> Matt Dainty <m...@bodgit-n-scarper.com> wrote:
> 
> > --- sys/arch/amd64/amd64/identcpu.c.orig    2013-01-14 22:23:43.000000000 
> > +0000
> > +++ sys/arch/amd64/amd64/identcpu.c 2013-01-14 22:33:21.000000000 +0000
> > @@ -506,7 +506,8 @@
> >             if (ci->ci_feature_sefflags & SEFF0EBX_SMAP)
> >                     replacesmap();
> >     }
> > -   if (!strncmp(mycpu_model, "Intel", 5)) {
> > +   if (!strncmp(mycpu_model, "Intel", 5) ||
> > +       !strncmp(mycpu_model, "Genuine Intel", 13)) {
> >             u_int32_t cflushsz;
> >  
> >             CPUID(0x01, dummy, cflushsz, dummy, dummy);
> > 
> 
> I think it's dubious that we match the CPU brand name for this at
> all.  Shouldn't this be properly handled with CPUID?

Here's a slightly simpler diff that checks the vendor rather than the
CPU brand.

I'm not sure if it's safe to check CPUID regardless of vendor, plus
there is the CLFLUSH code in that block as well, I figure that's
confined to Intel for a reason?

Matt

--- sys/arch/amd64/amd64/identcpu.c.orig        2013-01-14 22:23:43.000000000 
+0000
+++ sys/arch/amd64/amd64/identcpu.c     2013-01-29 22:17:44.000000000 +0000
@@ -506,7 +506,7 @@
                if (ci->ci_feature_sefflags & SEFF0EBX_SMAP)
                        replacesmap();
        }
-       if (!strncmp(mycpu_model, "Intel", 5)) {
+       if (!strcmp(cpu_vendor, "GenuineIntel")) {
                u_int32_t cflushsz;
 
                CPUID(0x01, dummy, cflushsz, dummy, dummy);