From: hch@lst.de (Christoph Hellwig) Date: Tue, 9 Oct 2018 08:13:43 +0200 Subject: [PATCH 1/2] nvme: NUMA locality information for fabrics In-Reply-To: References: <20181005092915.126636-1-hare@suse.de> <20181005092915.126636-2-hare@suse.de> <20181008100423.GA6686@lst.de> Message-ID: <20181009061343.GB6752@lst.de> On Mon, Oct 08, 2018@12:22:26PM +0200, Matias Bj?rling wrote: >> This changes behavior as node might remain NUMA_NO_NODE when the code >> above sets the device node to first_memory_node now when >> it would otherwise be NUMA_NO_NODE. Then again that code makes no >> sense to me to start with - Matias added it with the blk-mq conversion, >> so maybe he remembers why we do it to start with? If we want to fix >> up nodes like that it seems like we should do it in the driver core >> or the PCI core. >> > > Back in the days it was carried over to make sure the structures was > allocated from the same node as the pcie device. Well, the PCI device will only get allocated from a node if it has a node set, so overriding it doesn't make much sense. Now it seems like various architectures have a stub pcibus_to_node and thus always return NUMA_NO_NODE, but I don't really see a good reason to override it. Either we live with that, or move the set_dev_node hack into the PCI core.