On Wed, May 30, 2012 at 07:42:27AM -0700, H. Peter Anvin wrote:

On 05/30/2012 06:10 AM, Andre Przywara wrote:

quoted

Because we are behind a family check before tweaking the topology
bit, we can use the standard rd/wrmsr variants for the CPUID feature
register.

That is not what the *msr*_amd*() functions do.

NAK.  This is a totally bogus patch.

The *msr*_amd*() variants were used instead of the normal *msrl_safe
variants although the AMD variants weren't needed there at all.

This has no issue on baremetal but breaks xen and this is how we caught
this.

So the patch corrects the original patch so that xen is happy too.

-- 
Regards/Gruss,
Boris.