cvs commit: src/sys/amd64/amd64 identcpu.c src/sys/i386/i386 identcpu.c

[Jung-uk Kim]

On Monday 17 October 2005 06:00 pm, Brooks Davis wrote:
> On Tue, Oct 18, 2005 at 01:36:35AM +0400, Gleb Smirnoff wrote:
> >   David,
> >
> > On Mon, Oct 17, 2005 at 12:20:18PM -0700, David O'Brien wrote:
> > D> Other than my electric company no one (people or
> > D> FreeBSD SW) should care that I have dual-core CPU's.
> >
> > I don't agree. The dmesg output should describe the physical
> > layout of the hardware, too. When maintaining hundreds of systems
> > in server room, people usually forget what lives inside the
> > computer cases. And when one needs to tell what is the
> > motherboard and how many sockets and fans it has, he should be
> > capable to do this without opening the case.
>
> I'm not sure that I care what dmesg says, but I do want an accurate
> view of our topology presented by the system, not some arbitrary
> and frankly meaningless CPU count.  If nothing else I care because
> the two cores share one memory controller.  That may not be all
> that noticable now for normal application, but I expect it will be
> once we start seeing 4+ core CPUs.  Telling the user the (somewhat
> complicated) truth about their hardware is part of "doing it
> right".

I believe we have to implement ACPI SRAT (Static Resource Affinity 
Table) and SLIT (System Locality Information Table) to achieve this.  
Linux already does this for i386, amd64, and i64:

http://lxr.linux.no/source/arch/i386/kernel/srat.c
http://lxr.linux.no/source/arch/x86_64/mm/srat.c
http://lxr.linux.no/source/arch/ia64/kernel/acpi.c

Jung-uk Kim