* Christoph Lameter <cl@xxxxxxxxxxxxxxxxxxxx>: > On Wed, 13 Jan 2010, Alex Chiang wrote: > > > Firmware puts each cell into a NUMA node, so we should really > > only have 2 nodes, but for some reason, that 3rd node gets > > created too. I haven't inspected the SRAT/SLIT on this machine > > recently, but can do so if you want me to. > > May not have anything to do with the problem we are looking at but memory > setup is screwed up. Funky effects may follow. Actually, I was reminded off-list by my HP colleagues that the memory setup I showed you is common on mid-range and high-end HP ia64 platforms. Lee tells me: The third node is the "interleaved memory" pseudo-node. The firmware always interleaves 512MB of phys address space across the nodes. On these platforms, only interleaved memory is at phys addr 0--needed by firmware, ... All the real NUMA nodes' memory starts at some high phys addr. So, even in "numa mode" [a.k.a. 100% cell local memory], the firmware must create a region of interleaved memory at phys 0. So, we get N+1 nodes. Because node 2 is at phys 0 and contains only 512MB, it is all ZONE_DMA memory. DMA zone is 1st 4G on ia64. Our platforms have been shipping like this for years, so it's not like anything recent has changed. Thanks, /ac -- To unsubscribe from this list: send the line "unsubscribe linux-ia64" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
|