On Thu, Jul 07, 2016 at 09:30:19PM +0200, Alexander Gordeev wrote:

On Mon, Jul 04, 2016 at 05:39:33PM +0900, Christoph Hellwig wrote:

quoted

@@ -1575,6 +1546,7 @@ static int nvme_dev_add(struct nvme_dev *dev)
 		dev->tagset.cmd_size = nvme_cmd_size(dev);
 		dev->tagset.flags = BLK_MQ_F_SHOULD_MERGE;
 		dev->tagset.driver_data = dev;
+		dev->tagset.affinity_mask = to_pci_dev(dev->dev)->irq_affinity;
 
 		if (blk_mq_alloc_tag_set(&dev->tagset))
 			return 0;

Are there any post-init uses of blk_mq_tag_set::affinity_mask other than
calling to blk_mq_alloc_tag_set()? If no, blk_mq_tag_set::affinity_mask
is redundant, since the mask could be passed as a parameter.

We'll have to look at it in the block code when reinitializing rebuilding
the queue topology.  This isn't currently done, but we'll need it rather
soon.