Question about Opteron UnitID register

[Eric W. Biederman]

"David Hendricks" <sc at flagen.com> writes:

> I know you guys aren't AMD tech support, I was hoping someone could help
> me figure this problem out. In section 3.3.8 (page 35) of their BIOS and
> Kernel developers guide (
> http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/26094.PDF
> 
> ), they have a 32-bit UnitID register. The example given lists the lowest
> four bits as follows:
> 3-2: CPU1 Unit ID
> 1-2: CPU0 Unit ID
> 
> Ok, so you could give CPU1, CPU2, and CPU3 a unique ID with two binary
> digits, but I thought we were supposed to be able to have eight (or more?)
> CPUs?
> 
> Is there something obvious I'm missing? 
The context.  The unitid referred to are all internal to the cpu.

On current cpu implementations only CP0 is present so it is a little
confusing.

> Is the CPU Unit ID the same as the NodeID, or different? 
Different.

> Do these CPU Unit IDs even have to be unique for each CPU?

Yes.

> I was thinking perhaps the CPU1 Unit ID is used only by a noncoherant
> device attached via Hypertransport link to what it thinks is CPU1 to
> differentiate the CPU it's attached to from CPU0. If this is true, is it
> necessary since the bootstrap processor, CPU0, is the only one with access
> to its northbridge and memory controller at this point?
> 
> Sorry if all this is complete and utter nonsense, I'm just a very curious
> newbie.


There is a lot integrated into an opteron and it takes some serious looking
to really see what is going on.

Eric