Re: [Patch3/4]: fake numa for x86_64 patches

[Andi Kleen <ak@suse.de>]

On Monday 27 November 2006 18:59, Rohit Seth wrote:
> On Thu, 2006-11-23 at 10:04 +0100, Andi Kleen wrote:
> > On Wed, Nov 22, 2006 at 05:34:47PM -0800, Rohit Seth wrote:
> > > Fix the existing numa=fake so that ioholes are appropriately configured.
> > > Currently machines that have sizeable IO holes don't work with
> > > numa=fake>4.  This patch tries to equally partition the total available
> > > memory in equal size chunk.  The minimum size of the fake node is set to
> > > 32MB.
> > 
> > This patch seems to do far more than advertised in the change log? 
> > 
> > You're conflicting badly with Amul's numa hash function rewrite for example.
> > 
> 
> Both of these patches are mucking with hash function and
> populate_memnodemap.  I like Amul's approach of doing dynamic allocation
> of numa hash map so that it can support >64GB of memory space.  I will
> resend the patches on top of his patch (incorporating your other
> feedback).

FYI I dropped Amul's patch temporarily because it causes boot failures on 
some systems. But it will be likely readded once that problem is fixed.

-Andi