Re: AMD EPYC Topology problems

From: Thomas Gleixner
Date: Mon Dec 10 2018 - 16:52:35 EST


On Sun, 9 Dec 2018, Borislav Petkov wrote:

> On Mon, Dec 03, 2018 at 11:23:49AM +0000, Andrew Cooper wrote:
> > Right, but the documentation also states that where it says package, it
> > means "Node" in AMD's terminology, and the information in CPUID is per
> > socket, not per node.
> >
> > My point is that the numbers ending up in cpuinfo_x86 don't match the
> > semantics described by the documentation.
>
> Ok, I think I know where the issue stems from:
>
> definition of "package" in the AMD docs != definition of "package" in Documentation/x86/topology.txt
>
> AMD's is "Processor: A package containing one or more Nodes." whereas
> ours is:
>
> "Packages contain a number of cores plus shared resources, e.g. DRAM
> controller, shared caches etc."
>
> and physical sockets we don't care about because they're not relevant to
> sw.

Right. Physical sockets are not interesting at all.

> Yeah, lemme discuss this with tglx to refresh what we were thinking then. :)

What Intel calls package is called Node on AMD. Yes, it's a mess, but we
can't do much about it and as lots of existing code already used 'package'
for both Intel and AMD, there was no point to invent yet another name for
it. Maybe we should have...

Thanks,

tglx