Messages in this thread | | | Date | Fri, 19 Nov 2004 15:18:35 -0600 | From | Jack Steiner <> | Subject | SLIT and IO-only nodes |
| |
An SGI Altix SSI system consists of a collection of nodes connected via a high speed interconnect. Nodes come in several flavors:
- memory, IO and cpus - memory & cpus - memory only - IO only - (other combinations don't exist yet)
The first 2 types of nodes are typical.
You can think of the last 2 types of nodes as nodes that have been partially depopulated.
We need to describe all these nodes in the SLIT table. For example, when allocating memory on a memory-only node, knowing the distance to the node is important.
When assigning cpus to service interrupts for IO nodes or when creating driver memory structures for devices on IO nodes, it is important to use the nearest node that has cpus & memory.
On IA64, memory-only nodes are described in the SRAT, have a proximity domain number, NIDs, and appear in the SLIT. (ie., we don't have a problem with memory-only nodes).
However, IO-only nodes (AFAICT) cannot be described in the SLIT. The SLIT is indexed by proximity_domain_number (PXM). Currently, there is no SRAT entry for IO-only nodes. These nodes do not appear in the SLIT. It would seem that a new ACPI table is needed for IO-only nodes. The SRAT would describe the node & identify the IO buses that are attached to the node. I think this would give us what we need.
Before I start digging into the ACPI spec, has anyone already addressed this problem? Is this the right approach to take to solve the problem?
-- Thanks
Jack Steiner (steiner@sgi.com) 651-683-5302 Principal Engineer SGI - Silicon Graphics, Inc.
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |