Messages in this thread | | | Date | Wed, 23 Jan 2008 14:15:01 +0000 | From | Mel Gorman <> | Subject | Re: [PATCH 0/2] Relax restrictions on setting CONFIG_NUMA on x86 II |
| |
On (23/01/08 14:48), Andi Kleen didst pronounce: > On Wednesday 23 January 2008 12:24:36 Mel Gorman wrote: > > On (23/01/08 12:15), Andi Kleen didst pronounce: > > > Anyways from your earlier comments it sounds like you're trying to add > > > SRAT parsing to CONFIG_NUMAQ. Since that's redundant with the old > > > implementation it doesn't sound like a very useful thing to do. > > > > No, that would not be useful at all as it's redundant as you point out. The > > only reason to add it is if the Opteron box can figure out the CPU-to-node > > affinity. > > Assuming srat_32.c was fixed to not crash on Opteron it would likely > do that already without further changes. >
Understood.
> > :| The patches applied so far are about increasing test coverage, not SRAT > > messing. > > Test coverage of the NUMAQ kernel? >
NUMA in general. I don't really care about NUMAQ as such except that it continues to shake out the occasional bug that can be difficult to reproduce elsewhere.
> If you wanted to increase test coverage of 32bit NUMA kernels the right > strategy would be to fix srat_32. >
I will try and do that then instead of trying to merge the SRAT parsers. Based on this thread, my understanding is that an attempted merge would only open up a can of hurt, probably causing regressions in the process.
-- Mel Gorman Part-time Phd Student Linux Technology Center University of Limerick IBM Dublin Software Lab
| |