nvme-rdma: implement polling queue map

When passed with nr_poll_queues setup additional queues with cq polling
context IB_POLL_DIRECT (no interrupts) and make sure to set
QUEUE_FLAG_POLL on the connect_q. In addition add the third queue
mapping for polling queues.

nvmf connect on this queue is polled for like all other requests so make
nvmf_connect_io_queue poll for polling queues.
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>

drivers/nvme/host/rdma.c

@@ -162,6 +162,13 @@ static inline int nvme_rdma_queue_idx(struct nvme_rdma_queue *queue)
 	return queue - queue->ctrl->queues;
 }
 
+static bool nvme_rdma_poll_queue(struct nvme_rdma_queue *queue)
+{
+	return nvme_rdma_queue_idx(queue) >
+		queue->ctrl->ctrl.opts->nr_io_queues +
+		queue->ctrl->ctrl.opts->nr_write_queues;
+}
+
 static inline size_t nvme_rdma_inline_data_size(struct nvme_rdma_queue *queue)
 {
 	return queue->cmnd_capsule_len - sizeof(struct nvme_command);
@@ -440,6 +447,7 @@ static int nvme_rdma_create_queue_ib(struct nvme_rdma_queue *queue)
 	const int send_wr_factor = 3;			/* MR, SEND, INV */
 	const int cq_factor = send_wr_factor + 1;	/* + RECV */
 	int comp_vector, idx = nvme_rdma_queue_idx(queue);
+	enum ib_poll_context poll_ctx;
 	int ret;
 
 	queue->device = nvme_rdma_find_get_device(queue->cm_id);
@@ -456,10 +464,16 @@ static int nvme_rdma_create_queue_ib(struct nvme_rdma_queue *queue)
 	 */
 	comp_vector = idx == 0 ? idx : idx - 1;
 
+	/* Polling queues need direct cq polling context */
+	if (nvme_rdma_poll_queue(queue))
+		poll_ctx = IB_POLL_DIRECT;
+	else
+		poll_ctx = IB_POLL_SOFTIRQ;
+
 	/* +1 for ib_stop_cq */
 	queue->ib_cq = ib_alloc_cq(ibdev, queue,
 				cq_factor * queue->queue_size + 1,
-				comp_vector, IB_POLL_SOFTIRQ);
+				comp_vector, poll_ctx);
 	if (IS_ERR(queue->ib_cq)) {
 		ret = PTR_ERR(queue->ib_cq);
 		goto out_put_dev;
@@ -595,15 +609,17 @@ static void nvme_rdma_stop_io_queues(struct nvme_rdma_ctrl *ctrl)
 
 static int nvme_rdma_start_queue(struct nvme_rdma_ctrl *ctrl, int idx)
 {
+	struct nvme_rdma_queue *queue = &ctrl->queues[idx];
+	bool poll = nvme_rdma_poll_queue(queue);
 	int ret;
 
 	if (idx)
-		ret = nvmf_connect_io_queue(&ctrl->ctrl, idx, false);
+		ret = nvmf_connect_io_queue(&ctrl->ctrl, idx, poll);
 	else
 		ret = nvmf_connect_admin_queue(&ctrl->ctrl);
 
 	if (!ret)
-		set_bit(NVME_RDMA_Q_LIVE, &ctrl->queues[idx].flags);
+		set_bit(NVME_RDMA_Q_LIVE, &queue->flags);
 	else
 		dev_info(ctrl->ctrl.device,
 			"failed to connect queue: %d ret=%d\n", idx, ret);
@@ -646,6 +662,7 @@ static int nvme_rdma_alloc_io_queues(struct nvme_rdma_ctrl *ctrl)
 				ibdev->num_comp_vectors);
 
 	nr_io_queues += min(opts->nr_write_queues, num_online_cpus());
+	nr_io_queues += min(opts->nr_poll_queues, num_online_cpus());
 
 	ret = nvme_set_queue_count(&ctrl->ctrl, &nr_io_queues);
 	if (ret)
@@ -716,7 +733,7 @@ static struct blk_mq_tag_set *nvme_rdma_alloc_tagset(struct nvme_ctrl *nctrl,
 		set->driver_data = ctrl;
 		set->nr_hw_queues = nctrl->queue_count - 1;
 		set->timeout = NVME_IO_TIMEOUT;
-		set->nr_maps = 2 /* default + read */;
+		set->nr_maps = nctrl->opts->nr_poll_queues ? HCTX_MAX_TYPES : 2;
 	}
 
 	ret = blk_mq_alloc_tag_set(set);
@@ -1742,6 +1759,13 @@ static blk_status_t nvme_rdma_queue_rq(struct blk_mq_hw_ctx *hctx,
 	return BLK_STS_IOERR;
 }
 
+static int nvme_rdma_poll(struct blk_mq_hw_ctx *hctx)
+{
+	struct nvme_rdma_queue *queue = hctx->driver_data;
+
+	return ib_process_cq_direct(queue->ib_cq, -1);
+}
+
 static void nvme_rdma_complete_rq(struct request *rq)
 {
 	struct nvme_rdma_request *req = blk_mq_rq_to_pdu(rq);
@@ -1772,6 +1796,17 @@ static int nvme_rdma_map_queues(struct blk_mq_tag_set *set)
 			ctrl->device->dev, 0);
 	blk_mq_rdma_map_queues(&set->map[HCTX_TYPE_READ],
 			ctrl->device->dev, 0);
+
+	if (ctrl->ctrl.opts->nr_poll_queues) {
+		set->map[HCTX_TYPE_POLL].nr_queues =
+				ctrl->ctrl.opts->nr_poll_queues;
+		set->map[HCTX_TYPE_POLL].queue_offset =
+				ctrl->ctrl.opts->nr_io_queues;
+		if (ctrl->ctrl.opts->nr_write_queues)
+			set->map[HCTX_TYPE_POLL].queue_offset +=
+				ctrl->ctrl.opts->nr_write_queues;
+		blk_mq_map_queues(&set->map[HCTX_TYPE_POLL]);
+	}
 	return 0;
 }
 
@@ -1783,6 +1818,7 @@ static const struct blk_mq_ops nvme_rdma_mq_ops = {
 	.init_hctx	= nvme_rdma_init_hctx,
 	.timeout	= nvme_rdma_timeout,
 	.map_queues	= nvme_rdma_map_queues,
+	.poll		= nvme_rdma_poll,
 };
 
 static const struct blk_mq_ops nvme_rdma_admin_mq_ops = {
@@ -1927,7 +1963,8 @@ static struct nvme_ctrl *nvme_rdma_create_ctrl(struct device *dev,
 	INIT_WORK(&ctrl->err_work, nvme_rdma_error_recovery_work);
 	INIT_WORK(&ctrl->ctrl.reset_work, nvme_rdma_reset_ctrl_work);
 
-	ctrl->ctrl.queue_count = opts->nr_io_queues + opts->nr_write_queues + 1;
+	ctrl->ctrl.queue_count = opts->nr_io_queues + opts->nr_write_queues +
+				opts->nr_poll_queues + 1;
 	ctrl->ctrl.sqsize = opts->queue_size - 1;
 	ctrl->ctrl.kato = opts->kato;
 
@@ -1979,7 +2016,7 @@ static struct nvmf_transport_ops nvme_rdma_transport = {
 	.required_opts	= NVMF_OPT_TRADDR,
 	.allowed_opts	= NVMF_OPT_TRSVCID | NVMF_OPT_RECONNECT_DELAY |
 			  NVMF_OPT_HOST_TRADDR | NVMF_OPT_CTRL_LOSS_TMO |
-			  NVMF_OPT_NR_WRITE_QUEUES,
+			  NVMF_OPT_NR_WRITE_QUEUES | NVMF_OPT_NR_POLL_QUEUES,
 	.create_ctrl	= nvme_rdma_create_ctrl,
 };