From: hch@lst.de (Christoph Hellwig) Date: Thu, 11 Aug 2016 17:21:54 +0200 Subject: [PATCH RFC 2/4] nvme-rdma: fix sqsize/hsqsize/hrqsize per spec In-Reply-To: <1470888438-823-3-git-send-email-james_p_freyensee@linux.intel.com> References: <1470888438-823-1-git-send-email-james_p_freyensee@linux.intel.com> <1470888438-823-3-git-send-email-james_p_freyensee@linux.intel.com> Message-ID: <20160811152154.GB19738@lst.de> I think having different notations for hsqsize vs hrqsize is highly confusing, and I've asked for a clarification of the hrqsize definition which is ambious at it's best at the moment. But independent of that a few more comments: > @@ -1907,7 +1920,7 @@ static struct nvme_ctrl *nvme_rdma_create_ctrl(struct device *dev, > spin_lock_init(&ctrl->lock); > > ctrl->queue_count = opts->nr_io_queues + 1; /* +1 for admin queue */ > - ctrl->ctrl.sqsize = opts->queue_size; > + ctrl->ctrl.sqsize = opts->queue_size-1; We should keep our own sqsize in normal notation in core, and just convert to strange notations on the wire. Especially as we use it to driver all the resource allocation as Sagi said. > + if (priv.qid == 0) { > + priv.hsqsize = cpu_to_le16(queue->ctrl->ctrl.admin_sqsize); > + priv.hrqsize = cpu_to_le16(queue->ctrl->ctrl.admin_sqsize+1); Based on that priv.hsqsize should have a - 1 here and hrqsize should be kept as-is. Also always use spaces around the operators. And while we're at it - the fix to use the separate AQ values should go into the first patch.