[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Booting the kernel on very large NUMA systems



On Fri, 2014-01-10 at 09:59 +0100, John Paul Adrian Glaubitz wrote:
> On 01/09/2014 05:53 PM, John Paul Adrian Glaubitz wrote:
> > Ok, it's not an hardware issue. The SLES kernel (3.0 something) works
> > just fine:
> > 
> > [   26.877781] Brought up 1024 CPUs
> > [   26.881876] Total of 1024 processors activated (4642849.30 BogoMIPS).
> 
> The SLES and RHEL kernels have CONFIG_X86_UV enabled to support the APIC
> configuration of the SGI Altix UV1000. Debian doesn't which explains it
> fails to boot more than one physical CPU.
> 
> I'll have to roll my own kernel or use the one from SuSE or Red Hat.

Maybe we could turn this on for sid if you're actually interested in
running a Debian kernel on that machine.  It adds about 40K to the
(uncompressed) size of the kernel, though some of that will be discarded
after boot.

We would also need to increase CONFIG_NR_CPUS to support this machine
(currently set to 512 on amd64).  That also has a memory cost as some
arrays are sized based on this static value and not the actual number of
CPUs a given system supports.  I didn't check how much it is yet.

Ben.

-- 
Ben Hutchings
Always try to do things in chronological order;
it's less confusing that way.

Attachment: signature.asc
Description: This is a digitally signed message part


Reply to: