Messages in this thread | | | Date | Thu, 3 Mar 2011 12:04:01 -0800 (PST) | From | David Rientjes <> | Subject | Re: [PATCH x86/mm UPDATED] x86-64, NUMA: Fix distance table handling |
| |
On Wed, 2 Mar 2011, Yinghai Lu wrote:
> BTW: this conf get broken several times, and get fixed several times. >
Yes, it has been. I think it's because there's an assumption that if the NUMA initialization fails that we automatically fake a single node 0 covering all RAM. The pxm for node 0 doesn't actually need to have any memory bound to it in the SRAT, so it is trivial with a custom BIOS to avoid onlining the node. There doesn't appear to be any other problems related to this that we run into throughout the kernel, but it certainly needs to be fixed in this new emulation code even if those fake-to-phys mappings aren't currently used because it's a bad assumption.
| |