4 files changed, 155 insertions, 63 deletions

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 1b7c2afd84cb..120fb593d1da 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -3400,7 +3400,8 @@ static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
 {
 	struct nvme_ns *ns;
 	__le32 *ns_list;
-	unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
+	unsigned i, j, nsid, prev = 0;
+	unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
 	int ret = 0;
 
 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index f562154551ce..524d6bd6d095 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -2513,6 +2513,12 @@ static void nvme_reset_work(struct work_struct *work)
 	 */
 	dev->ctrl.max_hw_sectors = NVME_MAX_KB_SZ << 1;
 	dev->ctrl.max_segments = NVME_MAX_SEGS;
+
+	/*
+	 * Don't limit the IOMMU merged segment size.
+	 */
+	dma_set_max_seg_size(dev->dev, 0xffffffff);
+
 	mutex_unlock(&dev->shutdown_lock);
 
 	/*
diff --git a/drivers/nvme/host/rdma.c b/drivers/nvme/host/rdma.c
index f383146e7d0f..97f668a39ae1 100644
--- a/drivers/nvme/host/rdma.c
+++ b/drivers/nvme/host/rdma.c
@@ -213,6 +213,11 @@ static struct nvme_rdma_qe *nvme_rdma_alloc_ring(struct ib_device *ibdev,
 	if (!ring)
 		return NULL;
 
+	/*
+	 * Bind the CQEs (post recv buffers) DMA mapping to the RDMA queue
+	 * lifetime. It's safe, since any chage in the underlying RDMA device
+	 * will issue error recovery and queue re-creation.
+	 */
 	for (i = 0; i < ib_queue_size; i++) {
 		if (nvme_rdma_alloc_qe(ibdev, &ring[i], capsule_size, dir))
 			goto out_free_ring;
@@ -274,14 +279,9 @@ static int nvme_rdma_create_qp(struct nvme_rdma_queue *queue, const int factor)
 static void nvme_rdma_exit_request(struct blk_mq_tag_set *set,
 		struct request *rq, unsigned int hctx_idx)
 {
-	struct nvme_rdma_ctrl *ctrl = set->driver_data;
 	struct nvme_rdma_request *req = blk_mq_rq_to_pdu(rq);
-	int queue_idx = (set == &ctrl->tag_set) ? hctx_idx + 1 : 0;
-	struct nvme_rdma_queue *queue = &ctrl->queues[queue_idx];
-	struct nvme_rdma_device *dev = queue->device;
 
-	nvme_rdma_free_qe(dev->dev, &req->sqe, sizeof(struct nvme_command),
-			DMA_TO_DEVICE);
+	kfree(req->sqe.data);
 }
 
 static int nvme_rdma_init_request(struct blk_mq_tag_set *set,
@@ -292,15 +292,11 @@ static int nvme_rdma_init_request(struct blk_mq_tag_set *set,
 	struct nvme_rdma_request *req = blk_mq_rq_to_pdu(rq);
 	int queue_idx = (set == &ctrl->tag_set) ? hctx_idx + 1 : 0;
 	struct nvme_rdma_queue *queue = &ctrl->queues[queue_idx];
-	struct nvme_rdma_device *dev = queue->device;
-	struct ib_device *ibdev = dev->dev;
-	int ret;
 
 	nvme_req(rq)->ctrl = &ctrl->ctrl;
-	ret = nvme_rdma_alloc_qe(ibdev, &req->sqe, sizeof(struct nvme_command),
-			DMA_TO_DEVICE);
-	if (ret)
-		return ret;
+	req->sqe.data = kzalloc(sizeof(struct nvme_command), GFP_KERNEL);
+	if (!req->sqe.data)
+		return -ENOMEM;
 
 	req->queue = queue;
 
@@ -641,34 +637,16 @@ static int nvme_rdma_alloc_io_queues(struct nvme_rdma_ctrl *ctrl)
 {
 	struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
 	struct ib_device *ibdev = ctrl->device->dev;
-	unsigned int nr_io_queues;
+	unsigned int nr_io_queues, nr_default_queues;
+	unsigned int nr_read_queues, nr_poll_queues;
 	int i, ret;
 
-	nr_io_queues = min(opts->nr_io_queues, num_online_cpus());
-
-	/*
-	 * we map queues according to the device irq vectors for
-	 * optimal locality so we don't need more queues than
-	 * completion vectors.
-	 */
-	nr_io_queues = min_t(unsigned int, nr_io_queues,
-				ibdev->num_comp_vectors);
-
-	if (opts->nr_write_queues) {
-		ctrl->io_queues[HCTX_TYPE_DEFAULT] =
-				min(opts->nr_write_queues, nr_io_queues);
-		nr_io_queues += ctrl->io_queues[HCTX_TYPE_DEFAULT];
-	} else {
-		ctrl->io_queues[HCTX_TYPE_DEFAULT] = nr_io_queues;
-	}
-
-	ctrl->io_queues[HCTX_TYPE_READ] = nr_io_queues;
-
-	if (opts->nr_poll_queues) {
-		ctrl->io_queues[HCTX_TYPE_POLL] =
-			min(opts->nr_poll_queues, num_online_cpus());
-		nr_io_queues += ctrl->io_queues[HCTX_TYPE_POLL];
-	}
+	nr_read_queues = min_t(unsigned int, ibdev->num_comp_vectors,
+				min(opts->nr_io_queues, num_online_cpus()));
+	nr_default_queues =  min_t(unsigned int, ibdev->num_comp_vectors,
+				min(opts->nr_write_queues, num_online_cpus()));
+	nr_poll_queues = min(opts->nr_poll_queues, num_online_cpus());
+	nr_io_queues = nr_read_queues + nr_default_queues + nr_poll_queues;
 
 	ret = nvme_set_queue_count(&ctrl->ctrl, &nr_io_queues);
 	if (ret)
@@ -681,6 +659,34 @@ static int nvme_rdma_alloc_io_queues(struct nvme_rdma_ctrl *ctrl)
 	dev_info(ctrl->ctrl.device,
 		"creating %d I/O queues.\n", nr_io_queues);
 
+	if (opts->nr_write_queues && nr_read_queues < nr_io_queues) {
+		/*
+		 * separate read/write queues
+		 * hand out dedicated default queues only after we have
+		 * sufficient read queues.
+		 */
+		ctrl->io_queues[HCTX_TYPE_READ] = nr_read_queues;
+		nr_io_queues -= ctrl->io_queues[HCTX_TYPE_READ];
+		ctrl->io_queues[HCTX_TYPE_DEFAULT] =
+			min(nr_default_queues, nr_io_queues);
+		nr_io_queues -= ctrl->io_queues[HCTX_TYPE_DEFAULT];
+	} else {
+		/*
+		 * shared read/write queues
+		 * either no write queues were requested, or we don't have
+		 * sufficient queue count to have dedicated default queues.
+		 */
+		ctrl->io_queues[HCTX_TYPE_DEFAULT] =
+			min(nr_read_queues, nr_io_queues);
+		nr_io_queues -= ctrl->io_queues[HCTX_TYPE_DEFAULT];
+	}
+
+	if (opts->nr_poll_queues && nr_io_queues) {
+		/* map dedicated poll queues only if we have queues left */
+		ctrl->io_queues[HCTX_TYPE_POLL] =
+			min(nr_poll_queues, nr_io_queues);
+	}
+
 	for (i = 1; i < ctrl->ctrl.queue_count; i++) {
 		ret = nvme_rdma_alloc_queue(ctrl, i,
 				ctrl->ctrl.sqsize + 1);
@@ -769,6 +775,11 @@ static int nvme_rdma_configure_admin_queue(struct nvme_rdma_ctrl *ctrl,
 
 	ctrl->max_fr_pages = nvme_rdma_get_max_fr_pages(ctrl->device->dev);
 
+	/*
+	 * Bind the async event SQE DMA mapping to the admin queue lifetime.
+	 * It's safe, since any chage in the underlying RDMA device will issue
+	 * error recovery and queue re-creation.
+	 */
 	error = nvme_rdma_alloc_qe(ctrl->device->dev, &ctrl->async_event_sqe,
 			sizeof(struct nvme_command), DMA_TO_DEVICE);
 	if (error)
@@ -1709,12 +1720,20 @@ static blk_status_t nvme_rdma_queue_rq(struct blk_mq_hw_ctx *hctx,
 		return nvmf_fail_nonready_command(&queue->ctrl->ctrl, rq);
 
 	dev = queue->device->dev;
+
+	req->sqe.dma = ib_dma_map_single(dev, req->sqe.data,
+					 sizeof(struct nvme_command),
+					 DMA_TO_DEVICE);
+	err = ib_dma_mapping_error(dev, req->sqe.dma);
+	if (unlikely(err))
+		return BLK_STS_RESOURCE;
+
 	ib_dma_sync_single_for_cpu(dev, sqe->dma,
 			sizeof(struct nvme_command), DMA_TO_DEVICE);
 
 	ret = nvme_setup_cmd(ns, rq, c);
 	if (ret)
-		return ret;
+		goto unmap_qe;
 
 	blk_mq_start_request(rq);
 
@@ -1739,10 +1758,16 @@ static blk_status_t nvme_rdma_queue_rq(struct blk_mq_hw_ctx *hctx,
 	}
 
 	return BLK_STS_OK;
+
 err:
 	if (err == -ENOMEM || err == -EAGAIN)
-		return BLK_STS_RESOURCE;
-	return BLK_STS_IOERR;
+		ret = BLK_STS_RESOURCE;
+	else
+		ret = BLK_STS_IOERR;
+unmap_qe:
+	ib_dma_unmap_single(dev, req->sqe.dma, sizeof(struct nvme_command),
+			    DMA_TO_DEVICE);
+	return ret;
 }
 
 static int nvme_rdma_poll(struct blk_mq_hw_ctx *hctx)
@@ -1755,25 +1780,36 @@ static int nvme_rdma_poll(struct blk_mq_hw_ctx *hctx)
 static void nvme_rdma_complete_rq(struct request *rq)
 {
 	struct nvme_rdma_request *req = blk_mq_rq_to_pdu(rq);
+	struct nvme_rdma_queue *queue = req->queue;
+	struct ib_device *ibdev = queue->device->dev;
 
-	nvme_rdma_unmap_data(req->queue, rq);
+	nvme_rdma_unmap_data(queue, rq);
+	ib_dma_unmap_single(ibdev, req->sqe.dma, sizeof(struct nvme_command),
+			    DMA_TO_DEVICE);
 	nvme_complete_rq(rq);
 }
 
 static int nvme_rdma_map_queues(struct blk_mq_tag_set *set)
 {
 	struct nvme_rdma_ctrl *ctrl = set->driver_data;
+	struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
 
-	set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
-	set->map[HCTX_TYPE_DEFAULT].nr_queues =
-			ctrl->io_queues[HCTX_TYPE_DEFAULT];
-	set->map[HCTX_TYPE_READ].nr_queues = ctrl->io_queues[HCTX_TYPE_READ];
-	if (ctrl->ctrl.opts->nr_write_queues) {
+	if (opts->nr_write_queues && ctrl->io_queues[HCTX_TYPE_READ]) {
 		/* separate read/write queues */
+		set->map[HCTX_TYPE_DEFAULT].nr_queues =
+			ctrl->io_queues[HCTX_TYPE_DEFAULT];
+		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
+		set->map[HCTX_TYPE_READ].nr_queues =
+			ctrl->io_queues[HCTX_TYPE_READ];
 		set->map[HCTX_TYPE_READ].queue_offset =
-				ctrl->io_queues[HCTX_TYPE_DEFAULT];
+			ctrl->io_queues[HCTX_TYPE_DEFAULT];
 	} else {
-		/* mixed read/write queues */
+		/* shared read/write queues */
+		set->map[HCTX_TYPE_DEFAULT].nr_queues =
+			ctrl->io_queues[HCTX_TYPE_DEFAULT];
+		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
+		set->map[HCTX_TYPE_READ].nr_queues =
+			ctrl->io_queues[HCTX_TYPE_DEFAULT];
 		set->map[HCTX_TYPE_READ].queue_offset = 0;
 	}
 	blk_mq_rdma_map_queues(&set->map[HCTX_TYPE_DEFAULT],
@@ -1781,16 +1817,22 @@ static int nvme_rdma_map_queues(struct blk_mq_tag_set *set)
 	blk_mq_rdma_map_queues(&set->map[HCTX_TYPE_READ],
 			ctrl->device->dev, 0);
 
-	if (ctrl->ctrl.opts->nr_poll_queues) {
+	if (opts->nr_poll_queues && ctrl->io_queues[HCTX_TYPE_POLL]) {
+		/* map dedicated poll queues only if we have queues left */
 		set->map[HCTX_TYPE_POLL].nr_queues =
 				ctrl->io_queues[HCTX_TYPE_POLL];
 		set->map[HCTX_TYPE_POLL].queue_offset =
-				ctrl->io_queues[HCTX_TYPE_DEFAULT];
-		if (ctrl->ctrl.opts->nr_write_queues)
-			set->map[HCTX_TYPE_POLL].queue_offset +=
-				ctrl->io_queues[HCTX_TYPE_READ];
+			ctrl->io_queues[HCTX_TYPE_DEFAULT] +
+			ctrl->io_queues[HCTX_TYPE_READ];
 		blk_mq_map_queues(&set->map[HCTX_TYPE_POLL]);
 	}
+
+	dev_info(ctrl->ctrl.device,
+		"mapped %d/%d/%d default/read/poll queues.\n",
+		ctrl->io_queues[HCTX_TYPE_DEFAULT],
+		ctrl->io_queues[HCTX_TYPE_READ],
+		ctrl->io_queues[HCTX_TYPE_POLL]);
+
 	return 0;
 }
 
diff --git a/drivers/nvme/host/tcp.c b/drivers/nvme/host/tcp.c
index 2b107a1d152b..08a2501b9357 100644
--- a/drivers/nvme/host/tcp.c
+++ b/drivers/nvme/host/tcp.c
@@ -111,6 +111,7 @@ struct nvme_tcp_ctrl {
 	struct work_struct	err_work;
 	struct delayed_work	connect_work;
 	struct nvme_tcp_request async_req;
+	u32			io_queues[HCTX_MAX_TYPES];
 };
 
 static LIST_HEAD(nvme_tcp_ctrl_list);
@@ -1564,6 +1565,35 @@ static unsigned int nvme_tcp_nr_io_queues(struct nvme_ctrl *ctrl)
 	return nr_io_queues;
 }
 
+static void nvme_tcp_set_io_queues(struct nvme_ctrl *nctrl,
+		unsigned int nr_io_queues)
+{
+	struct nvme_tcp_ctrl *ctrl = to_tcp_ctrl(nctrl);
+	struct nvmf_ctrl_options *opts = nctrl->opts;
+
+	if (opts->nr_write_queues && opts->nr_io_queues < nr_io_queues) {
+		/*
+		 * separate read/write queues
+		 * hand out dedicated default queues only after we have
+		 * sufficient read queues.
+		 */
+		ctrl->io_queues[HCTX_TYPE_READ] = opts->nr_io_queues;
+		nr_io_queues -= ctrl->io_queues[HCTX_TYPE_READ];
+		ctrl->io_queues[HCTX_TYPE_DEFAULT] =
+			min(opts->nr_write_queues, nr_io_queues);
+		nr_io_queues -= ctrl->io_queues[HCTX_TYPE_DEFAULT];
+	} else {
+		/*
+		 * shared read/write queues
+		 * either no write queues were requested, or we don't have
+		 * sufficient queue count to have dedicated default queues.
+		 */
+		ctrl->io_queues[HCTX_TYPE_DEFAULT] =
+			min(opts->nr_io_queues, nr_io_queues);
+		nr_io_queues -= ctrl->io_queues[HCTX_TYPE_DEFAULT];
+	}
+}
+
 static int nvme_tcp_alloc_io_queues(struct nvme_ctrl *ctrl)
 {
 	unsigned int nr_io_queues;
@@ -1581,6 +1611,8 @@ static int nvme_tcp_alloc_io_queues(struct nvme_ctrl *ctrl)
 	dev_info(ctrl->device,
 		"creating %d I/O queues.\n", nr_io_queues);
 
+	nvme_tcp_set_io_queues(ctrl, nr_io_queues);
+
 	return __nvme_tcp_alloc_io_queues(ctrl);
 }
 
@@ -2089,23 +2121,34 @@ static blk_status_t nvme_tcp_queue_rq(struct blk_mq_hw_ctx *hctx,
 static int nvme_tcp_map_queues(struct blk_mq_tag_set *set)
 {
 	struct nvme_tcp_ctrl *ctrl = set->driver_data;
+	struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
 
-	set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
-	set->map[HCTX_TYPE_READ].nr_queues = ctrl->ctrl.opts->nr_io_queues;
-	if (ctrl->ctrl.opts->nr_write_queues) {
+	if (opts->nr_write_queues && ctrl->io_queues[HCTX_TYPE_READ]) {
 		/* separate read/write queues */
 		set->map[HCTX_TYPE_DEFAULT].nr_queues =
-				ctrl->ctrl.opts->nr_write_queues;
+			ctrl->io_queues[HCTX_TYPE_DEFAULT];
+		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
+		set->map[HCTX_TYPE_READ].nr_queues =
+			ctrl->io_queues[HCTX_TYPE_READ];
 		set->map[HCTX_TYPE_READ].queue_offset =
-				ctrl->ctrl.opts->nr_write_queues;
+			ctrl->io_queues[HCTX_TYPE_DEFAULT];
 	} else {
-		/* mixed read/write queues */
+		/* shared read/write queues */
 		set->map[HCTX_TYPE_DEFAULT].nr_queues =
-				ctrl->ctrl.opts->nr_io_queues;
+			ctrl->io_queues[HCTX_TYPE_DEFAULT];
+		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
+		set->map[HCTX_TYPE_READ].nr_queues =
+			ctrl->io_queues[HCTX_TYPE_DEFAULT];
 		set->map[HCTX_TYPE_READ].queue_offset = 0;
 	}
 	blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
 	blk_mq_map_queues(&set->map[HCTX_TYPE_READ]);
+
+	dev_info(ctrl->ctrl.device,
+		"mapped %d/%d default/read queues.\n",
+		ctrl->io_queues[HCTX_TYPE_DEFAULT],
+		ctrl->io_queues[HCTX_TYPE_READ]);
+
 	return 0;
 }