On Wed, Aug 15, 2007 at 11:13:20PM +0200, Stefan Reinauer wrote:
- Robert Millan rmh@aybabtu.com [070815 15:00]:
I had to take this part of your patch to fix Opteron miss-detection.
Are you checking for the cpu itself? If this is so, I think CPUID would be much more reliable. Would you like a patch for that?
Is CPUID reliable? (I guess it is, but that might make a pretty complex function)
Please do send a patch. I think cpuid would be a good way to go.
CPUID is reliable I think, but I'm not sure what we need to test exactly:
- Test for Dual Opteron as well?
- Comments in the source code suggest that socket 939 Athlon64 also should match this check. This poses two problems:
- Uwe's patch precisely disabled the check because it was breaking boot on his s939 athlon64 (and it also does in mine). However, I suspect this could be related to the memory access problem Rudolf just sent a mail about.
- Some s939/athlon64 models cannot be distinguished properly with cpuid_eax(1) since the return value is shared with athlon64 fx or sempron. Does this indicate we need to match those? Or do we want to tell them appart in some other way? (processor name string should work for that, although strcmp'ing that has a feeling of is-not-quite-right over it)