iser-target: introduce fast memory registration mode (FRWR)

This model was introduced in 00f7ec36c "RDMA/core: Add memory
management extensions support" and works when the IB device
supports the IB_DEVICE_MEM_MGT_EXTENSIONS capability.

Upon creating the isert device, ib_isert will test whether the HCA
supports FRWR. If supported then set the flag and assign
function pointers that handle fast registration and deregistration
of appropriate resources (fast_reg descriptors).

When new connection coming in, ib_isert will check frwr flag and
create frwr resouces, if fail to do it will switch back to
old model of using global dma key and turn off the frwr support.

Registration is done using posting IB_WR_FAST_REG_MR to the QP and
invalidations using posting IB_WR_LOCAL_INV.

Signed-off-by: Vu Pham <vu@mellanox.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>

drivers/infiniband/ulp/isert/ib_isert.c

@@ -45,6 +45,11 @@ isert_unmap_cmd(struct isert_cmd *isert_cmd, struct isert_conn *isert_conn);
 static int
 isert_map_rdma(struct iscsi_conn *conn, struct iscsi_cmd *cmd,
 	       struct isert_rdma_wr *wr);
+static void
+isert_unreg_rdma_frwr(struct isert_cmd *isert_cmd, struct isert_conn *isert_conn);
+static int
+isert_reg_rdma_frwr(struct iscsi_conn *conn, struct iscsi_cmd *cmd,
+		    struct isert_rdma_wr *wr);
 
 static void
 isert_qp_event_callback(struct ib_event *e, void *context)
@@ -85,14 +90,8 @@ isert_conn_setup_qp(struct isert_conn *isert_conn, struct rdma_cm_id *cma_id)
 {
 	struct isert_device *device = isert_conn->conn_device;
 	struct ib_qp_init_attr attr;
-	struct ib_device_attr devattr;
 	int ret, index, min_index = 0;
 
-	memset(&devattr, 0, sizeof(struct ib_device_attr));
-	ret = isert_query_device(cma_id->device, &devattr);
-	if (ret)
-		return ret;
-
 	mutex_lock(&device_list_mutex);
 	for (index = 0; index < device->cqs_used; index++)
 		if (device->cq_active_qps[index] <
@@ -113,7 +112,7 @@ isert_conn_setup_qp(struct isert_conn *isert_conn, struct rdma_cm_id *cma_id)
 	 * FIXME: Use devattr.max_sge - 2 for max_send_sge as
 	 * work-around for RDMA_READ..
 	 */
-	attr.cap.max_send_sge = devattr.max_sge - 2;
+	attr.cap.max_send_sge = device->dev_attr.max_sge - 2;
 	isert_conn->max_sge = attr.cap.max_send_sge;
 
 	attr.cap.max_recv_sge = 1;
@@ -215,18 +214,31 @@ isert_create_device_ib_res(struct isert_device *device)
 {
 	struct ib_device *ib_dev = device->ib_device;
 	struct isert_cq_desc *cq_desc;
+	struct ib_device_attr *dev_attr;
 	int ret = 0, i, j;
 
+	dev_attr = &device->dev_attr;
+	ret = isert_query_device(ib_dev, dev_attr);
+	if (ret)
+		return ret;
+
 	/* asign function handlers */
-	device->reg_rdma_mem = isert_map_rdma;
-	device->unreg_rdma_mem = isert_unmap_cmd;
+	if (dev_attr->device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) {
+		device->use_frwr = 1;
+		device->reg_rdma_mem = isert_reg_rdma_frwr;
+		device->unreg_rdma_mem = isert_unreg_rdma_frwr;
+	} else {
+		device->use_frwr = 0;
+		device->reg_rdma_mem = isert_map_rdma;
+		device->unreg_rdma_mem = isert_unmap_cmd;
+	}
 
 	device->cqs_used = min_t(int, num_online_cpus(),
 				 device->ib_device->num_comp_vectors);
 	device->cqs_used = min(ISERT_MAX_CQ, device->cqs_used);
-	pr_debug("Using %d CQs, device %s supports %d vectors\n",
+	pr_debug("Using %d CQs, device %s supports %d vectors support FRWR %d\n",
 		 device->cqs_used, device->ib_device->name,
-		 device->ib_device->num_comp_vectors);
+		 device->ib_device->num_comp_vectors, device->use_frwr);
 	device->cq_desc = kzalloc(sizeof(struct isert_cq_desc) *
 				device->cqs_used, GFP_KERNEL);
 	if (!device->cq_desc) {
@@ -372,6 +384,85 @@ isert_device_find_by_ib_dev(struct rdma_cm_id *cma_id)
 	return device;
 }
 
+static void
+isert_conn_free_frwr_pool(struct isert_conn *isert_conn)
+{
+	struct fast_reg_descriptor *fr_desc, *tmp;
+	int i = 0;
+
+	if (list_empty(&isert_conn->conn_frwr_pool))
+		return;
+
+	pr_debug("Freeing conn %p frwr pool", isert_conn);
+
+	list_for_each_entry_safe(fr_desc, tmp,
+				 &isert_conn->conn_frwr_pool, list) {
+		list_del(&fr_desc->list);
+		ib_free_fast_reg_page_list(fr_desc->data_frpl);
+		ib_dereg_mr(fr_desc->data_mr);
+		kfree(fr_desc);
+		++i;
+	}
+
+	if (i < isert_conn->conn_frwr_pool_size)
+		pr_warn("Pool still has %d regions registered\n",
+			isert_conn->conn_frwr_pool_size - i);
+}
+
+static int
+isert_conn_create_frwr_pool(struct isert_conn *isert_conn)
+{
+	struct fast_reg_descriptor *fr_desc;
+	struct isert_device *device = isert_conn->conn_device;
+	int i, ret;
+
+	INIT_LIST_HEAD(&isert_conn->conn_frwr_pool);
+	isert_conn->conn_frwr_pool_size = 0;
+	for (i = 0; i < ISCSI_DEF_XMIT_CMDS_MAX; i++) {
+		fr_desc = kzalloc(sizeof(*fr_desc), GFP_KERNEL);
+		if (!fr_desc) {
+			pr_err("Failed to allocate fast_reg descriptor\n");
+			ret = -ENOMEM;
+			goto err;
+		}
+
+		fr_desc->data_frpl =
+			ib_alloc_fast_reg_page_list(device->ib_device,
+						    ISCSI_ISER_SG_TABLESIZE);
+		if (IS_ERR(fr_desc->data_frpl)) {
+			pr_err("Failed to allocate fr_pg_list err=%ld\n",
+			       PTR_ERR(fr_desc->data_frpl));
+			ret = PTR_ERR(fr_desc->data_frpl);
+			goto err;
+		}
+
+		fr_desc->data_mr = ib_alloc_fast_reg_mr(device->dev_pd,
+					ISCSI_ISER_SG_TABLESIZE);
+		if (IS_ERR(fr_desc->data_mr)) {
+			pr_err("Failed to allocate frmr err=%ld\n",
+			       PTR_ERR(fr_desc->data_mr));
+			ret = PTR_ERR(fr_desc->data_mr);
+			ib_free_fast_reg_page_list(fr_desc->data_frpl);
+			goto err;
+		}
+		pr_debug("Create fr_desc %p page_list %p\n",
+			 fr_desc, fr_desc->data_frpl->page_list);
+
+		fr_desc->valid = true;
+		list_add_tail(&fr_desc->list, &isert_conn->conn_frwr_pool);
+		isert_conn->conn_frwr_pool_size++;
+	}
+
+	pr_debug("Creating conn %p frwr pool size=%d",
+		 isert_conn, isert_conn->conn_frwr_pool_size);
+
+	return 0;
+
+err:
+	isert_conn_free_frwr_pool(isert_conn);
+	return ret;
+}
+
 static int
 isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
 {
@@ -398,6 +489,7 @@ isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
 	kref_init(&isert_conn->conn_kref);
 	kref_get(&isert_conn->conn_kref);
 	mutex_init(&isert_conn->conn_mutex);
+	spin_lock_init(&isert_conn->conn_lock);
 
 	cma_id->context = isert_conn;
 	isert_conn->conn_cm_id = cma_id;
@@ -455,6 +547,14 @@ isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
 	isert_conn->conn_pd = device->dev_pd;
 	isert_conn->conn_mr = device->dev_mr;
 
+	if (device->use_frwr) {
+		ret = isert_conn_create_frwr_pool(isert_conn);
+		if (ret) {
+			pr_err("Conn: %p failed to create frwr_pool\n", isert_conn);
+			goto out_frwr;
+		}
+	}
+
 	ret = isert_conn_setup_qp(isert_conn, cma_id);
 	if (ret)
 		goto out_conn_dev;
@@ -468,6 +568,9 @@ isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
 	return 0;
 
 out_conn_dev:
+	if (device->use_frwr)
+		isert_conn_free_frwr_pool(isert_conn);
+out_frwr:
 	isert_device_try_release(device);
 out_rsp_dma_map:
 	ib_dma_unmap_single(ib_dev, isert_conn->login_rsp_dma,
@@ -491,6 +594,9 @@ isert_connect_release(struct isert_conn *isert_conn)
 
 	pr_debug("Entering isert_connect_release(): >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>\n");
 
+	if (device->use_frwr)
+		isert_conn_free_frwr_pool(isert_conn);
+
 	if (isert_conn->conn_qp) {
 		cq_index = ((struct isert_cq_desc *)
 			isert_conn->conn_qp->recv_cq->cq_context)->cq_index;
@@ -1265,6 +1371,36 @@ isert_unmap_cmd(struct isert_cmd *isert_cmd, struct isert_conn *isert_conn)
 	}
 }
 
+static void
+isert_unreg_rdma_frwr(struct isert_cmd *isert_cmd, struct isert_conn *isert_conn)
+{
+	struct isert_rdma_wr *wr = &isert_cmd->rdma_wr;
+	struct ib_device *ib_dev = isert_conn->conn_cm_id->device;
+	LIST_HEAD(unmap_list);
+
+	pr_debug("unreg_frwr_cmd: %p\n", isert_cmd);
+
+	if (wr->fr_desc) {
+		pr_debug("unreg_frwr_cmd: %p free fr_desc %p\n",
+			 isert_cmd, wr->fr_desc);
+		spin_lock_bh(&isert_conn->conn_lock);
+		list_add_tail(&wr->fr_desc->list, &isert_conn->conn_frwr_pool);
+		spin_unlock_bh(&isert_conn->conn_lock);
+		wr->fr_desc = NULL;
+	}
+
+	if (wr->sge) {
+		pr_debug("unreg_frwr_cmd: %p unmap_sg op\n", isert_cmd);
+		ib_dma_unmap_sg(ib_dev, wr->sge, wr->num_sge,
+				(wr->iser_ib_op == ISER_IB_RDMA_WRITE) ?
+				DMA_TO_DEVICE : DMA_FROM_DEVICE);
+		wr->sge = NULL;
+	}
+
+	wr->ib_sge = NULL;
+	wr->send_wr = NULL;
+}
+
 static void
 isert_put_cmd(struct isert_cmd *isert_cmd)
 {
@@ -1940,12 +2076,217 @@ unmap_sg:
 	return ret;
 }
 
+static int
+isert_map_fr_pagelist(struct ib_device *ib_dev,
+		      struct scatterlist *sg_start, int sg_nents, u64 *fr_pl)
+{
+	u64 start_addr, end_addr, page, chunk_start = 0;
+	struct scatterlist *tmp_sg;
+	int i = 0, new_chunk, last_ent, n_pages;
+
+	n_pages = 0;
+	new_chunk = 1;
+	last_ent = sg_nents - 1;
+	for_each_sg(sg_start, tmp_sg, sg_nents, i) {
+		start_addr = ib_sg_dma_address(ib_dev, tmp_sg);
+		if (new_chunk)
+			chunk_start = start_addr;
+		end_addr = start_addr + ib_sg_dma_len(ib_dev, tmp_sg);
+
+		pr_debug("SGL[%d] dma_addr: 0x%16llx len: %u\n",
+			 i, (unsigned long long)tmp_sg->dma_address,
+			 tmp_sg->length);
+
+		if ((end_addr & ~PAGE_MASK) && i < last_ent) {
+			new_chunk = 0;
+			continue;
+		}
+		new_chunk = 1;
+
+		page = chunk_start & PAGE_MASK;
+		do {
+			fr_pl[n_pages++] = page;
+			pr_debug("Mapped page_list[%d] page_addr: 0x%16llx\n",
+				 n_pages - 1, page);
+			page += PAGE_SIZE;
+		} while (page < end_addr);
+	}
+
+	return n_pages;
+}
+
+static int
+isert_fast_reg_mr(struct fast_reg_descriptor *fr_desc,
+		  struct isert_cmd *isert_cmd, struct isert_conn *isert_conn,
+		  struct ib_sge *ib_sge, u32 offset, unsigned int data_len)
+{
+	struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd;
+	struct ib_device *ib_dev = isert_conn->conn_cm_id->device;
+	struct scatterlist *sg_start;
+	u32 sg_off, page_off;
+	struct ib_send_wr fr_wr, inv_wr;
+	struct ib_send_wr *bad_wr, *wr = NULL;
+	u8 key;
+	int ret, sg_nents, pagelist_len;
+
+	sg_off = offset / PAGE_SIZE;
+	sg_start = &cmd->se_cmd.t_data_sg[sg_off];
+	sg_nents = min_t(unsigned int, cmd->se_cmd.t_data_nents - sg_off,
+			 ISCSI_ISER_SG_TABLESIZE);
+	page_off = offset % PAGE_SIZE;
+
+	pr_debug("Cmd: %p use fr_desc %p sg_nents %d sg_off %d offset %u\n",
+		 isert_cmd, fr_desc, sg_nents, sg_off, offset);
+
+	pagelist_len = isert_map_fr_pagelist(ib_dev, sg_start, sg_nents,
+					     &fr_desc->data_frpl->page_list[0]);
+
+	if (!fr_desc->valid) {
+		memset(&inv_wr, 0, sizeof(inv_wr));
+		inv_wr.opcode = IB_WR_LOCAL_INV;
+		inv_wr.ex.invalidate_rkey = fr_desc->data_mr->rkey;
+		wr = &inv_wr;
+		/* Bump the key */
+		key = (u8)(fr_desc->data_mr->rkey & 0x000000FF);
+		ib_update_fast_reg_key(fr_desc->data_mr, ++key);
+	}
+
+	/* Prepare FASTREG WR */
+	memset(&fr_wr, 0, sizeof(fr_wr));
+	fr_wr.opcode = IB_WR_FAST_REG_MR;
+	fr_wr.wr.fast_reg.iova_start =
+		fr_desc->data_frpl->page_list[0] + page_off;
+	fr_wr.wr.fast_reg.page_list = fr_desc->data_frpl;
+	fr_wr.wr.fast_reg.page_list_len = pagelist_len;
+	fr_wr.wr.fast_reg.page_shift = PAGE_SHIFT;
+	fr_wr.wr.fast_reg.length = data_len;
+	fr_wr.wr.fast_reg.rkey = fr_desc->data_mr->rkey;
+	fr_wr.wr.fast_reg.access_flags = IB_ACCESS_LOCAL_WRITE;
+
+	if (!wr)
+		wr = &fr_wr;
+	else
+		wr->next = &fr_wr;
+
+	ret = ib_post_send(isert_conn->conn_qp, wr, &bad_wr);
+	if (ret) {
+		pr_err("fast registration failed, ret:%d\n", ret);
+		return ret;
+	}
+	fr_desc->valid = false;
+
+	ib_sge->lkey = fr_desc->data_mr->lkey;
+	ib_sge->addr = fr_desc->data_frpl->page_list[0] + page_off;
+	ib_sge->length = data_len;
+
+	pr_debug("RDMA ib_sge: addr: 0x%16llx  length: %u lkey: %08x\n",
+		 ib_sge->addr, ib_sge->length, ib_sge->lkey);
+
+	return ret;
+}
+
+static int
+isert_reg_rdma_frwr(struct iscsi_conn *conn, struct iscsi_cmd *cmd,
+		    struct isert_rdma_wr *wr)
+{
+	struct se_cmd *se_cmd = &cmd->se_cmd;
+	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
+	struct isert_conn *isert_conn = (struct isert_conn *)conn->context;
+	struct ib_device *ib_dev = isert_conn->conn_cm_id->device;
+	struct ib_send_wr *send_wr;
+	struct ib_sge *ib_sge;
+	struct scatterlist *sg_start;
+	struct fast_reg_descriptor *fr_desc;
+	u32 sg_off = 0, sg_nents;
+	u32 offset = 0, data_len, data_left, rdma_write_max;
+	int ret = 0, count;
+	unsigned long flags;
+
+	if (wr->iser_ib_op == ISER_IB_RDMA_WRITE) {
+		data_left = se_cmd->data_length;
+		iscsit_increment_maxcmdsn(cmd, conn->sess);
+		cmd->stat_sn = conn->stat_sn++;
+	} else {
+		sg_off = cmd->write_data_done / PAGE_SIZE;
+		data_left = se_cmd->data_length - cmd->write_data_done;
+		offset = cmd->write_data_done;
+		isert_cmd->tx_desc.isert_cmd = isert_cmd;
+	}
+
+	sg_start = &cmd->se_cmd.t_data_sg[sg_off];
+	sg_nents = se_cmd->t_data_nents - sg_off;
+
+	count = ib_dma_map_sg(ib_dev, sg_start, sg_nents,
+			      (wr->iser_ib_op == ISER_IB_RDMA_WRITE) ?
+			      DMA_TO_DEVICE : DMA_FROM_DEVICE);
+	if (unlikely(!count)) {
+		pr_err("Cmd: %p unrable to map SGs\n", isert_cmd);
+		return -EINVAL;
+	}
+	wr->sge = sg_start;
+	wr->num_sge = sg_nents;
+	pr_debug("Mapped cmd: %p count: %u sg: %p sg_nents: %u rdma_len %d\n",
+		 isert_cmd, count, sg_start, sg_nents, data_left);
+
+	memset(&wr->s_ib_sge, 0, sizeof(*ib_sge));
+	ib_sge = &wr->s_ib_sge;
+	wr->ib_sge = ib_sge;
+
+	wr->send_wr_num = 1;
+	memset(&wr->s_send_wr, 0, sizeof(*send_wr));
+	wr->send_wr = &wr->s_send_wr;
+
+	wr->isert_cmd = isert_cmd;
+	rdma_write_max = ISCSI_ISER_SG_TABLESIZE * PAGE_SIZE;
+
+	send_wr = &isert_cmd->rdma_wr.s_send_wr;
+	send_wr->sg_list = ib_sge;
+	send_wr->num_sge = 1;
+	send_wr->wr_id = (unsigned long)&isert_cmd->tx_desc;
+	if (wr->iser_ib_op == ISER_IB_RDMA_WRITE) {
+		send_wr->opcode = IB_WR_RDMA_WRITE;
+		send_wr->wr.rdma.remote_addr = isert_cmd->read_va;
+		send_wr->wr.rdma.rkey = isert_cmd->read_stag;
+		send_wr->send_flags = 0;
+		send_wr->next = &isert_cmd->tx_desc.send_wr;
+	} else {
+		send_wr->opcode = IB_WR_RDMA_READ;
+		send_wr->wr.rdma.remote_addr = isert_cmd->write_va;
+		send_wr->wr.rdma.rkey = isert_cmd->write_stag;
+		send_wr->send_flags = IB_SEND_SIGNALED;
+	}
+
+	data_len = min(data_left, rdma_write_max);
+	wr->cur_rdma_length = data_len;
+
+	spin_lock_irqsave(&isert_conn->conn_lock, flags);
+	fr_desc = list_first_entry(&isert_conn->conn_frwr_pool,
+				   struct fast_reg_descriptor, list);
+	list_del(&fr_desc->list);
+	spin_unlock_irqrestore(&isert_conn->conn_lock, flags);
+	wr->fr_desc = fr_desc;
+
+	ret = isert_fast_reg_mr(fr_desc, isert_cmd, isert_conn,
+			  ib_sge, offset, data_len);
+	if (ret) {
+		list_add_tail(&fr_desc->list, &isert_conn->conn_frwr_pool);
+		goto unmap_sg;
+	}
+
+	return 0;
+
+unmap_sg:
+	ib_dma_unmap_sg(ib_dev, sg_start, sg_nents,
+			(wr->iser_ib_op == ISER_IB_RDMA_WRITE) ?
+			DMA_TO_DEVICE : DMA_FROM_DEVICE);
+	return ret;
+}
+
 static int
 isert_put_datain(struct iscsi_conn *conn, struct iscsi_cmd *cmd)
 {
 	struct se_cmd *se_cmd = &cmd->se_cmd;
-	struct isert_cmd *isert_cmd = container_of(cmd,
-					struct isert_cmd, iscsi_cmd);
+	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
 	struct isert_rdma_wr *wr = &isert_cmd->rdma_wr;
 	struct isert_conn *isert_conn = (struct isert_conn *)conn->context;
 	struct isert_device *device = isert_conn->conn_device;

drivers/infiniband/ulp/isert/ib_isert.h

@@ -5,6 +5,7 @@
 #include <rdma/rdma_cm.h>
 
 #define ISERT_RDMA_LISTEN_BACKLOG	10
+#define ISCSI_ISER_SG_TABLESIZE		256
 
 enum isert_desc_type {
 	ISCSI_TX_CONTROL,
@@ -45,16 +46,26 @@ struct iser_tx_desc {
 	struct ib_send_wr send_wr;
 } __packed;
 
+struct fast_reg_descriptor {
+	struct list_head	list;
+	struct ib_mr		*data_mr;
+	struct ib_fast_reg_page_list	*data_frpl;
+	bool			valid;
+};
+
 struct isert_rdma_wr {
 	struct list_head	wr_list;
 	struct isert_cmd	*isert_cmd;
 	enum iser_ib_op_code	iser_ib_op;
 	struct ib_sge		*ib_sge;
+	struct ib_sge		s_ib_sge;
 	int			num_sge;
 	struct scatterlist	*sge;
 	int			send_wr_num;
 	struct ib_send_wr	*send_wr;
+	struct ib_send_wr	s_send_wr;
 	u32			cur_rdma_length;
+	struct fast_reg_descriptor *fr_desc;
 };
 
 struct isert_cmd {
@@ -106,6 +117,10 @@ struct isert_conn {
 	wait_queue_head_t	conn_wait;
 	wait_queue_head_t	conn_wait_comp_err;
 	struct kref		conn_kref;
+	struct list_head	conn_frwr_pool;
+	int			conn_frwr_pool_size;
+	/* lock to protect frwr_pool */
+	spinlock_t		conn_lock;
 };
 
 #define ISERT_MAX_CQ 64
@@ -118,6 +133,7 @@ struct isert_cq_desc {
 };
 
 struct isert_device {
+	int			use_frwr;
 	int			cqs_used;
 	int			refcount;
 	int			cq_active_qps[ISERT_MAX_CQ];
@@ -128,6 +144,7 @@ struct isert_device {
 	struct ib_cq		*dev_tx_cq[ISERT_MAX_CQ];
 	struct isert_cq_desc	*cq_desc;
 	struct list_head	dev_node;
+	struct ib_device_attr	dev_attr;
 	int			(*reg_rdma_mem)(struct iscsi_conn *conn,
 						    struct iscsi_cmd *cmd,
 						    struct isert_rdma_wr *wr);