Merge branch 'iser' into for-next

drivers/infiniband/ulp/iser/iscsi_iser.c

@@ -128,6 +128,28 @@ static int iscsi_iser_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
 	return 0;
 }
 
+int iser_initialize_task_headers(struct iscsi_task *task,
+						struct iser_tx_desc *tx_desc)
+{
+	struct iscsi_iser_conn *iser_conn = task->conn->dd_data;
+	struct iser_device     *device    = iser_conn->ib_conn->device;
+	struct iscsi_iser_task *iser_task = task->dd_data;
+	u64 dma_addr;
+
+	dma_addr = ib_dma_map_single(device->ib_device, (void *)tx_desc,
+				ISER_HEADERS_LEN, DMA_TO_DEVICE);
+	if (ib_dma_mapping_error(device->ib_device, dma_addr))
+		return -ENOMEM;
+
+	tx_desc->dma_addr = dma_addr;
+	tx_desc->tx_sg[0].addr   = tx_desc->dma_addr;
+	tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
+	tx_desc->tx_sg[0].lkey   = device->mr->lkey;
+
+	iser_task->headers_initialized	= 1;
+	iser_task->iser_conn		= iser_conn;
+	return 0;
+}
 /**
  * iscsi_iser_task_init - Initialize task
  * @task: iscsi task
@@ -137,17 +159,17 @@ static int iscsi_iser_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
 static int
 iscsi_iser_task_init(struct iscsi_task *task)
 {
-	struct iscsi_iser_conn *iser_conn  = task->conn->dd_data;
 	struct iscsi_iser_task *iser_task = task->dd_data;
 
+	if (!iser_task->headers_initialized)
+		if (iser_initialize_task_headers(task, &iser_task->desc))
+			return -ENOMEM;
+
 	/* mgmt task */
-	if (!task->sc) {
-		iser_task->desc.data = task->data;
+	if (!task->sc)
 		return 0;
-	}
 
 	iser_task->command_sent = 0;
-	iser_task->iser_conn    = iser_conn;
 	iser_task_rdma_init(iser_task);
 	return 0;
 }
@@ -168,7 +190,7 @@ iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task)
 {
 	int error = 0;
 
-	iser_dbg("task deq [cid %d itt 0x%x]\n", conn->id, task->itt);
+	iser_dbg("mtask xmit [cid %d itt 0x%x]\n", conn->id, task->itt);
 
 	error = iser_send_control(conn, task);
 
@@ -178,9 +200,6 @@ iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task)
 	 * - if yes, the task is recycled at iscsi_complete_pdu
 	 * - if no,  the task is recycled at iser_snd_completion
 	 */
-	if (error && error != -ENOBUFS)
-		iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
-
 	return error;
 }
 
@@ -232,7 +251,7 @@ iscsi_iser_task_xmit(struct iscsi_task *task)
 			   task->imm_count, task->unsol_r2t.data_length);
 	}
 
-	iser_dbg("task deq [cid %d itt 0x%x]\n",
+	iser_dbg("ctask xmit [cid %d itt 0x%x]\n",
 		   conn->id, task->itt);
 
 	/* Send the cmd PDU */
@@ -248,8 +267,6 @@ iscsi_iser_task_xmit(struct iscsi_task *task)
 		error = iscsi_iser_task_xmit_unsol_data(conn, task);
 
  iscsi_iser_task_xmit_exit:
-	if (error && error != -ENOBUFS)
-		iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
 	return error;
 }
 
@@ -283,7 +300,7 @@ iscsi_iser_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
 	 * due to issues with the login code re iser sematics
 	 * this not set in iscsi_conn_setup - FIXME
 	 */
-	conn->max_recv_dlength = 128;
+	conn->max_recv_dlength = ISER_RECV_DATA_SEG_LEN;
 
 	iser_conn = conn->dd_data;
 	conn->dd_data = iser_conn;
@@ -401,7 +418,7 @@ iscsi_iser_session_create(struct iscsi_endpoint *ep,
 	struct Scsi_Host *shost;
 	struct iser_conn *ib_conn;
 
-	shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 1);
+	shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0);
 	if (!shost)
 		return NULL;
 	shost->transportt = iscsi_iser_scsi_transport;
@@ -675,7 +692,7 @@ static int __init iser_init(void)
 	memset(&ig, 0, sizeof(struct iser_global));
 
 	ig.desc_cache = kmem_cache_create("iser_descriptors",
-					  sizeof (struct iser_desc),
+					  sizeof(struct iser_tx_desc),
 					  0, SLAB_HWCACHE_ALIGN,
 					  NULL);
 	if (ig.desc_cache == NULL)

drivers/infiniband/ulp/iser/iscsi_iser.h

@@ -102,9 +102,9 @@
 #define ISER_MAX_TX_MISC_PDUS		6 /* NOOP_OUT(2), TEXT(1),         *
 					   * SCSI_TMFUNC(2), LOGOUT(1) */
 
-#define ISER_QP_MAX_RECV_DTOS		(ISCSI_DEF_XMIT_CMDS_MAX + \
-					ISER_MAX_RX_MISC_PDUS    +  \
-					ISER_MAX_TX_MISC_PDUS)
+#define ISER_QP_MAX_RECV_DTOS		(ISCSI_DEF_XMIT_CMDS_MAX)
+
+#define ISER_MIN_POSTED_RX		(ISCSI_DEF_XMIT_CMDS_MAX >> 2)
 
 /* the max TX (send) WR supported by the iSER QP is defined by                 *
  * max_send_wr = T * (1 + D) + C ; D is how many inflight dataouts we expect   *
@@ -132,6 +132,12 @@ struct iser_hdr {
 	__be64  read_va;
 } __attribute__((packed));
 
+/* Constant PDU lengths calculations */
+#define ISER_HEADERS_LEN  (sizeof(struct iser_hdr) + sizeof(struct iscsi_hdr))
+
+#define ISER_RECV_DATA_SEG_LEN	128
+#define ISER_RX_PAYLOAD_SIZE	(ISER_HEADERS_LEN + ISER_RECV_DATA_SEG_LEN)
+#define ISER_RX_LOGIN_SIZE	(ISER_HEADERS_LEN + ISCSI_DEF_MAX_RECV_SEG_LEN)
 
 /* Length of an object name string */
 #define ISER_OBJECT_NAME_SIZE		    64
@@ -187,51 +193,43 @@ struct iser_regd_buf {
 	struct iser_mem_reg     reg;        /* memory registration info        */
 	void                    *virt_addr;
 	struct iser_device      *device;    /* device->device for dma_unmap    */
-	u64                     dma_addr;   /* if non zero, addr for dma_unmap */
 	enum dma_data_direction direction;  /* direction for dma_unmap	       */
 	unsigned int            data_size;
-	atomic_t                ref_count;  /* refcount, freed when dec to 0   */
-};
-
-#define MAX_REGD_BUF_VECTOR_LEN	2
-
-struct iser_dto {
-	struct iscsi_iser_task *task;
-	struct iser_conn *ib_conn;
-	int                        notify_enable;
-
-	/* vector of registered buffers */
-	unsigned int               regd_vector_len;
-	struct iser_regd_buf       *regd[MAX_REGD_BUF_VECTOR_LEN];
-
-	/* offset into the registered buffer may be specified */
-	unsigned int               offset[MAX_REGD_BUF_VECTOR_LEN];
-
-	/* a smaller size may be specified, if 0, then full size is used */
-	unsigned int               used_sz[MAX_REGD_BUF_VECTOR_LEN];
 };
 
 enum iser_desc_type {
-	ISCSI_RX,
 	ISCSI_TX_CONTROL ,
 	ISCSI_TX_SCSI_COMMAND,
 	ISCSI_TX_DATAOUT
 };
 
-struct iser_desc {
+struct iser_tx_desc {
 	struct iser_hdr              iser_header;
 	struct iscsi_hdr             iscsi_header;
-	struct iser_regd_buf         hdr_regd_buf;
-	void                         *data;         /* used by RX & TX_CONTROL */
-	struct iser_regd_buf         data_regd_buf; /* used by RX & TX_CONTROL */
 	enum   iser_desc_type        type;
-	struct iser_dto              dto;
+	u64		             dma_addr;
+	/* sg[0] points to iser/iscsi headers, sg[1] optionally points to either
+	of immediate data, unsolicited data-out or control (login,text) */
+	struct ib_sge		     tx_sg[2];
+	int                          num_sge;
 };
 
+#define ISER_RX_PAD_SIZE	(256 - (ISER_RX_PAYLOAD_SIZE + \
+					sizeof(u64) + sizeof(struct ib_sge)))
+struct iser_rx_desc {
+	struct iser_hdr              iser_header;
+	struct iscsi_hdr             iscsi_header;
+	char		             data[ISER_RECV_DATA_SEG_LEN];
+	u64		             dma_addr;
+	struct ib_sge		     rx_sg;
+	char		             pad[ISER_RX_PAD_SIZE];
+} __attribute__((packed));
+
 struct iser_device {
 	struct ib_device             *ib_device;
 	struct ib_pd	             *pd;
-	struct ib_cq	             *cq;
+	struct ib_cq	             *rx_cq;
+	struct ib_cq	             *tx_cq;
 	struct ib_mr	             *mr;
 	struct tasklet_struct	     cq_tasklet;
 	struct list_head             ig_list; /* entry in ig devices list */
@@ -250,15 +248,18 @@ struct iser_conn {
 	struct ib_fmr_pool           *fmr_pool;     /* pool of IB FMRs         */
 	int                          disc_evt_flag; /* disconn event delivered */
 	wait_queue_head_t	     wait;          /* waitq for conn/disconn  */
-	atomic_t                     post_recv_buf_count; /* posted rx count   */
+	int                          post_recv_buf_count; /* posted rx count  */
 	atomic_t                     post_send_buf_count; /* posted tx count   */
-	atomic_t                     unexpected_pdu_count;/* count of received *
-							   * unexpected pdus   *
-							   * not yet retired   */
 	char 			     name[ISER_OBJECT_NAME_SIZE];
 	struct iser_page_vec         *page_vec;     /* represents SG to fmr maps*
 						     * maps serialized as tx is*/
 	struct list_head	     conn_list;       /* entry in ig conn list */
+
+	char  			     *login_buf;
+	u64 			     login_dma;
+	unsigned int 		     rx_desc_head;
+	struct iser_rx_desc	     *rx_descs;
+	struct ib_recv_wr	     rx_wr[ISER_MIN_POSTED_RX];
 };
 
 struct iscsi_iser_conn {
@@ -267,7 +268,7 @@ struct iscsi_iser_conn {
 };
 
 struct iscsi_iser_task {
-	struct iser_desc             desc;
+	struct iser_tx_desc          desc;
 	struct iscsi_iser_conn	     *iser_conn;
 	enum iser_task_status 	     status;
 	int                          command_sent;  /* set if command  sent  */
@@ -275,6 +276,7 @@ struct iscsi_iser_task {
 	struct iser_regd_buf         rdma_regd[ISER_DIRS_NUM];/* regd rdma buf */
 	struct iser_data_buf         data[ISER_DIRS_NUM];     /* orig. data des*/
 	struct iser_data_buf         data_copy[ISER_DIRS_NUM];/* contig. copy  */
+	int                          headers_initialized;
 };
 
 struct iser_page_vec {
@@ -322,22 +324,17 @@ void iser_conn_put(struct iser_conn *ib_conn);
 
 void iser_conn_terminate(struct iser_conn *ib_conn);
 
-void iser_rcv_completion(struct iser_desc *desc,
-			 unsigned long    dto_xfer_len);
+void iser_rcv_completion(struct iser_rx_desc *desc,
+			 unsigned long    dto_xfer_len,
+			struct iser_conn *ib_conn);
 
-void iser_snd_completion(struct iser_desc *desc);
+void iser_snd_completion(struct iser_tx_desc *desc, struct iser_conn *ib_conn);
 
 void iser_task_rdma_init(struct iscsi_iser_task *task);
 
 void iser_task_rdma_finalize(struct iscsi_iser_task *task);
 
-void iser_dto_buffs_release(struct iser_dto *dto);
-
-int  iser_regd_buff_release(struct iser_regd_buf *regd_buf);
-
-void iser_reg_single(struct iser_device      *device,
-		     struct iser_regd_buf    *regd_buf,
-		     enum dma_data_direction direction);
+void iser_free_rx_descriptors(struct iser_conn *ib_conn);
 
 void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *task,
 				     enum iser_data_dir         cmd_dir);
@@ -356,11 +353,9 @@ int  iser_reg_page_vec(struct iser_conn     *ib_conn,
 
 void iser_unreg_mem(struct iser_mem_reg *mem_reg);
 
-int  iser_post_recv(struct iser_desc *rx_desc);
-int  iser_post_send(struct iser_desc *tx_desc);
-
-int iser_conn_state_comp(struct iser_conn *ib_conn,
-			 enum iser_ib_conn_state comp);
+int  iser_post_recvl(struct iser_conn *ib_conn);
+int  iser_post_recvm(struct iser_conn *ib_conn, int count);
+int  iser_post_send(struct iser_conn *ib_conn, struct iser_tx_desc *tx_desc);
 
 int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
 			    struct iser_data_buf       *data,
@@ -368,4 +363,6 @@ int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
 			    enum   dma_data_direction  dma_dir);
 
 void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task);
+int  iser_initialize_task_headers(struct iscsi_task *task,
+			struct iser_tx_desc *tx_desc);
 #endif

drivers/infiniband/ulp/iser/iser_initiator.c

@@ -39,29 +39,6 @@
 
 #include "iscsi_iser.h"
 
-/* Constant PDU lengths calculations */
-#define ISER_TOTAL_HEADERS_LEN  (sizeof (struct iser_hdr) + \
-				 sizeof (struct iscsi_hdr))
-
-/* iser_dto_add_regd_buff - increments the reference count for *
- * the registered buffer & adds it to the DTO object           */
-static void iser_dto_add_regd_buff(struct iser_dto *dto,
-				   struct iser_regd_buf *regd_buf,
-				   unsigned long use_offset,
-				   unsigned long use_size)
-{
-	int add_idx;
-
-	atomic_inc(&regd_buf->ref_count);
-
-	add_idx = dto->regd_vector_len;
-	dto->regd[add_idx] = regd_buf;
-	dto->used_sz[add_idx] = use_size;
-	dto->offset[add_idx] = use_offset;
-
-	dto->regd_vector_len++;
-}
-
 /* Register user buffer memory and initialize passive rdma
  *  dto descriptor. Total data size is stored in
  *  iser_task->data[ISER_DIR_IN].data_len
@@ -122,9 +99,9 @@ iser_prepare_write_cmd(struct iscsi_task *task,
 	struct iscsi_iser_task *iser_task = task->dd_data;
 	struct iser_regd_buf *regd_buf;
 	int err;
-	struct iser_dto *send_dto = &iser_task->desc.dto;
 	struct iser_hdr *hdr = &iser_task->desc.iser_header;
 	struct iser_data_buf *buf_out = &iser_task->data[ISER_DIR_OUT];
+	struct ib_sge *tx_dsg = &iser_task->desc.tx_sg[1];
 
 	err = iser_dma_map_task_data(iser_task,
 				     buf_out,
@@ -163,135 +140,100 @@ iser_prepare_write_cmd(struct iscsi_task *task,
 	if (imm_sz > 0) {
 		iser_dbg("Cmd itt:%d, WRITE, adding imm.data sz: %d\n",
 			 task->itt, imm_sz);
-		iser_dto_add_regd_buff(send_dto,
-				       regd_buf,
-				       0,
-				       imm_sz);
+		tx_dsg->addr   = regd_buf->reg.va;
+		tx_dsg->length = imm_sz;
+		tx_dsg->lkey   = regd_buf->reg.lkey;
+		iser_task->desc.num_sge = 2;
 	}
 
 	return 0;
 }
 
-/**
- * iser_post_receive_control - allocates, initializes and posts receive DTO.
- */
-static int iser_post_receive_control(struct iscsi_conn *conn)
+/* creates a new tx descriptor and adds header regd buffer */
+static void iser_create_send_desc(struct iser_conn	*ib_conn,
+				  struct iser_tx_desc	*tx_desc)
 {
-	struct iscsi_iser_conn *iser_conn = conn->dd_data;
-	struct iser_desc     *rx_desc;
-	struct iser_regd_buf *regd_hdr;
-	struct iser_regd_buf *regd_data;
-	struct iser_dto      *recv_dto = NULL;
-	struct iser_device  *device = iser_conn->ib_conn->device;
-	int rx_data_size, err;
-	int posts, outstanding_unexp_pdus;
-
-	/* for the login sequence we must support rx of upto 8K; login is done
-	 * after conn create/bind (connect) and conn stop/bind (reconnect),
-	 * what's common for both schemes is that the connection is not started
-	 */
-	if (conn->c_stage != ISCSI_CONN_STARTED)
-		rx_data_size = ISCSI_DEF_MAX_RECV_SEG_LEN;
-	else /* FIXME till user space sets conn->max_recv_dlength correctly */
-		rx_data_size = 128;
+	struct iser_device *device = ib_conn->device;
 
-	outstanding_unexp_pdus =
-		atomic_xchg(&iser_conn->ib_conn->unexpected_pdu_count, 0);
+	ib_dma_sync_single_for_cpu(device->ib_device,
+		tx_desc->dma_addr, ISER_HEADERS_LEN, DMA_TO_DEVICE);
 
-	/*
-	 * in addition to the response buffer, replace those consumed by
-	 * unexpected pdus.
-	 */
-	for (posts = 0; posts < 1 + outstanding_unexp_pdus; posts++) {
-		rx_desc = kmem_cache_alloc(ig.desc_cache, GFP_NOIO);
-		if (rx_desc == NULL) {
-			iser_err("Failed to alloc desc for post recv %d\n",
-				 posts);
-			err = -ENOMEM;
-			goto post_rx_cache_alloc_failure;
-		}
-		rx_desc->type = ISCSI_RX;
-		rx_desc->data = kmalloc(rx_data_size, GFP_NOIO);
-		if (rx_desc->data == NULL) {
-			iser_err("Failed to alloc data buf for post recv %d\n",
-				 posts);
-			err = -ENOMEM;
-			goto post_rx_kmalloc_failure;
-		}
+	memset(&tx_desc->iser_header, 0, sizeof(struct iser_hdr));
+	tx_desc->iser_header.flags = ISER_VER;
 
-		recv_dto = &rx_desc->dto;
-		recv_dto->ib_conn = iser_conn->ib_conn;
-		recv_dto->regd_vector_len = 0;
+	tx_desc->num_sge = 1;
 
-		regd_hdr = &rx_desc->hdr_regd_buf;
-		memset(regd_hdr, 0, sizeof(struct iser_regd_buf));
-		regd_hdr->device  = device;
-		regd_hdr->virt_addr  = rx_desc; /* == &rx_desc->iser_header */
-		regd_hdr->data_size  = ISER_TOTAL_HEADERS_LEN;
+	if (tx_desc->tx_sg[0].lkey != device->mr->lkey) {
+		tx_desc->tx_sg[0].lkey = device->mr->lkey;
+		iser_dbg("sdesc %p lkey mismatch, fixing\n", tx_desc);
+	}
+}
 
-		iser_reg_single(device, regd_hdr, DMA_FROM_DEVICE);
 
-		iser_dto_add_regd_buff(recv_dto, regd_hdr, 0, 0);
+int iser_alloc_rx_descriptors(struct iser_conn *ib_conn)
+{
+	int i, j;
+	u64 dma_addr;
+	struct iser_rx_desc *rx_desc;
+	struct ib_sge       *rx_sg;
+	struct iser_device  *device = ib_conn->device;
 
-		regd_data = &rx_desc->data_regd_buf;
-		memset(regd_data, 0, sizeof(struct iser_regd_buf));
-		regd_data->device  = device;
-		regd_data->virt_addr  = rx_desc->data;
-		regd_data->data_size  = rx_data_size;
+	ib_conn->rx_descs = kmalloc(ISER_QP_MAX_RECV_DTOS *
+				sizeof(struct iser_rx_desc), GFP_KERNEL);
+	if (!ib_conn->rx_descs)
+		goto rx_desc_alloc_fail;
 
-		iser_reg_single(device, regd_data, DMA_FROM_DEVICE);
+	rx_desc = ib_conn->rx_descs;
 
-		iser_dto_add_regd_buff(recv_dto, regd_data, 0, 0);
+	for (i = 0; i < ISER_QP_MAX_RECV_DTOS; i++, rx_desc++)  {
+		dma_addr = ib_dma_map_single(device->ib_device, (void *)rx_desc,
+					ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
+		if (ib_dma_mapping_error(device->ib_device, dma_addr))
+			goto rx_desc_dma_map_failed;
 
-		err = iser_post_recv(rx_desc);
-		if (err) {
-			iser_err("Failed iser_post_recv for post %d\n", posts);
-			goto post_rx_post_recv_failure;
-		}
-	}
-	/* all posts successful */
-	return 0;
+		rx_desc->dma_addr = dma_addr;
 
-post_rx_post_recv_failure:
-	iser_dto_buffs_release(recv_dto);
-	kfree(rx_desc->data);
-post_rx_kmalloc_failure:
-	kmem_cache_free(ig.desc_cache, rx_desc);
-post_rx_cache_alloc_failure:
-	if (posts > 0) {
-		/*
-		 * response buffer posted, but did not replace all unexpected
-		 * pdu recv bufs. Ignore error, retry occurs next send
-		 */
-		outstanding_unexp_pdus -= (posts - 1);
-		err = 0;
+		rx_sg = &rx_desc->rx_sg;
+		rx_sg->addr   = rx_desc->dma_addr;
+		rx_sg->length = ISER_RX_PAYLOAD_SIZE;
+		rx_sg->lkey   = device->mr->lkey;
 	}
-	atomic_add(outstanding_unexp_pdus,
-		   &iser_conn->ib_conn->unexpected_pdu_count);
 
-	return err;
+	ib_conn->rx_desc_head = 0;
+	return 0;
+
+rx_desc_dma_map_failed:
+	rx_desc = ib_conn->rx_descs;
+	for (j = 0; j < i; j++, rx_desc++)
+		ib_dma_unmap_single(device->ib_device, rx_desc->dma_addr,
+			ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
+	kfree(ib_conn->rx_descs);
+	ib_conn->rx_descs = NULL;
+rx_desc_alloc_fail:
+	iser_err("failed allocating rx descriptors / data buffers\n");
+	return -ENOMEM;
 }
 
-/* creates a new tx descriptor and adds header regd buffer */
-static void iser_create_send_desc(struct iscsi_iser_conn *iser_conn,
-				  struct iser_desc       *tx_desc)
+void iser_free_rx_descriptors(struct iser_conn *ib_conn)
 {
-	struct iser_regd_buf *regd_hdr = &tx_desc->hdr_regd_buf;
-	struct iser_dto      *send_dto = &tx_desc->dto;
+	int i;
+	struct iser_rx_desc *rx_desc;
+	struct iser_device *device = ib_conn->device;
 
-	memset(regd_hdr, 0, sizeof(struct iser_regd_buf));
-	regd_hdr->device  = iser_conn->ib_conn->device;
-	regd_hdr->virt_addr  = tx_desc; /* == &tx_desc->iser_header */
-	regd_hdr->data_size  = ISER_TOTAL_HEADERS_LEN;
+	if (ib_conn->login_buf) {
+		ib_dma_unmap_single(device->ib_device, ib_conn->login_dma,
+			ISER_RX_LOGIN_SIZE, DMA_FROM_DEVICE);
+		kfree(ib_conn->login_buf);
+	}
 
-	send_dto->ib_conn         = iser_conn->ib_conn;
-	send_dto->notify_enable   = 1;
-	send_dto->regd_vector_len = 0;
+	if (!ib_conn->rx_descs)
+		return;
 
-	memset(&tx_desc->iser_header, 0, sizeof(struct iser_hdr));
-	tx_desc->iser_header.flags = ISER_VER;
-
-	iser_dto_add_regd_buff(send_dto, regd_hdr, 0, 0);
+	rx_desc = ib_conn->rx_descs;
+	for (i = 0; i < ISER_QP_MAX_RECV_DTOS; i++, rx_desc++)
+		ib_dma_unmap_single(device->ib_device, rx_desc->dma_addr,
+			ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
+	kfree(ib_conn->rx_descs);
 }
 
 /**
@@ -301,46 +243,23 @@ int iser_conn_set_full_featured_mode(struct iscsi_conn *conn)
 {
 	struct iscsi_iser_conn *iser_conn = conn->dd_data;
 
-	int i;
-	/*
-	 * FIXME this value should be declared to the target during login with
-	 * the MaxOutstandingUnexpectedPDUs key when supported
-	 */
-	int initial_post_recv_bufs_num = ISER_MAX_RX_MISC_PDUS;
-
-	iser_dbg("Initially post: %d\n", initial_post_recv_bufs_num);
+	iser_dbg("Initially post: %d\n", ISER_MIN_POSTED_RX);
 
 	/* Check that there is no posted recv or send buffers left - */
 	/* they must be consumed during the login phase */
-	BUG_ON(atomic_read(&iser_conn->ib_conn->post_recv_buf_count) != 0);
+	BUG_ON(iser_conn->ib_conn->post_recv_buf_count != 0);
 	BUG_ON(atomic_read(&iser_conn->ib_conn->post_send_buf_count) != 0);
 
-	/* Initial post receive buffers */
-	for (i = 0; i < initial_post_recv_bufs_num; i++) {
-		if (iser_post_receive_control(conn) != 0) {
-			iser_err("Failed to post recv bufs at:%d conn:0x%p\n",
-				 i, conn);
+	if (iser_alloc_rx_descriptors(iser_conn->ib_conn))
 		return -ENOMEM;
-		}
-	}
-	iser_dbg("Posted %d post recv bufs, conn:0x%p\n", i, conn);
-	return 0;
-}
 
-static int
-iser_check_xmit(struct iscsi_conn *conn, void *task)
-{
-	struct iscsi_iser_conn *iser_conn = conn->dd_data;
+	/* Initial post receive buffers */
+	if (iser_post_recvm(iser_conn->ib_conn, ISER_MIN_POSTED_RX))
+		return -ENOMEM;
 
-	if (atomic_read(&iser_conn->ib_conn->post_send_buf_count) ==
-	    ISER_QP_MAX_REQ_DTOS) {
-		iser_dbg("%ld can't xmit task %p\n",jiffies,task);
-		return -ENOBUFS;
-	}
 	return 0;
 }
 
-
 /**
  * iser_send_command - send command PDU
  */
@@ -349,27 +268,18 @@ int iser_send_command(struct iscsi_conn *conn,
 {
 	struct iscsi_iser_conn *iser_conn = conn->dd_data;
 	struct iscsi_iser_task *iser_task = task->dd_data;
-	struct iser_dto *send_dto = NULL;
 	unsigned long edtl;
-	int err = 0;
+	int err;
 	struct iser_data_buf *data_buf;
 	struct iscsi_cmd *hdr =  (struct iscsi_cmd *)task->hdr;
 	struct scsi_cmnd *sc  =  task->sc;
-
-	if (!iser_conn_state_comp(iser_conn->ib_conn, ISER_CONN_UP)) {
-		iser_err("Failed to send, conn: 0x%p is not up\n", iser_conn->ib_conn);
-		return -EPERM;
-	}
-	if (iser_check_xmit(conn, task))
-		return -ENOBUFS;
+	struct iser_tx_desc *tx_desc = &iser_task->desc;
 
 	edtl = ntohl(hdr->data_length);
 
 	/* build the tx desc regd header and add it to the tx desc dto */
-	iser_task->desc.type = ISCSI_TX_SCSI_COMMAND;
-	send_dto = &iser_task->desc.dto;
-	send_dto->task = iser_task;
-	iser_create_send_desc(iser_conn, &iser_task->desc);
+	tx_desc->type = ISCSI_TX_SCSI_COMMAND;
+	iser_create_send_desc(iser_conn->ib_conn, tx_desc);
 
 	if (hdr->flags & ISCSI_FLAG_CMD_READ)
 		data_buf = &iser_task->data[ISER_DIR_IN];
@@ -398,23 +308,13 @@ int iser_send_command(struct iscsi_conn *conn,
 			goto send_command_error;
 	}
 
-	iser_reg_single(iser_conn->ib_conn->device,
-			send_dto->regd[0], DMA_TO_DEVICE);
-
-	if (iser_post_receive_control(conn) != 0) {
-		iser_err("post_recv failed!\n");
-		err = -ENOMEM;
-		goto send_command_error;
-	}
-
 	iser_task->status = ISER_TASK_STATUS_STARTED;
 
-	err = iser_post_send(&iser_task->desc);
+	err = iser_post_send(iser_conn->ib_conn, tx_desc);
 	if (!err)
 		return 0;
 
 send_command_error:
-	iser_dto_buffs_release(send_dto);
 	iser_err("conn %p failed task->itt %d err %d\n",conn, task->itt, err);
 	return err;
 }
@@ -428,20 +328,13 @@ int iser_send_data_out(struct iscsi_conn *conn,
 {
 	struct iscsi_iser_conn *iser_conn = conn->dd_data;
 	struct iscsi_iser_task *iser_task = task->dd_data;
-	struct iser_desc *tx_desc = NULL;
-	struct iser_dto *send_dto = NULL;
+	struct iser_tx_desc *tx_desc = NULL;
+	struct iser_regd_buf *regd_buf;
 	unsigned long buf_offset;
 	unsigned long data_seg_len;
 	uint32_t itt;
 	int err = 0;
-
-	if (!iser_conn_state_comp(iser_conn->ib_conn, ISER_CONN_UP)) {
-		iser_err("Failed to send, conn: 0x%p is not up\n", iser_conn->ib_conn);
-		return -EPERM;
-	}
-
-	if (iser_check_xmit(conn, task))
-		return -ENOBUFS;
+	struct ib_sge *tx_dsg;
 
 	itt = (__force uint32_t)hdr->itt;
 	data_seg_len = ntoh24(hdr->dlength);
@@ -450,28 +343,25 @@ int iser_send_data_out(struct iscsi_conn *conn,
 	iser_dbg("%s itt %d dseg_len %d offset %d\n",
 		 __func__,(int)itt,(int)data_seg_len,(int)buf_offset);
 
-	tx_desc = kmem_cache_alloc(ig.desc_cache, GFP_NOIO);
+	tx_desc = kmem_cache_zalloc(ig.desc_cache, GFP_ATOMIC);
 	if (tx_desc == NULL) {
 		iser_err("Failed to alloc desc for post dataout\n");
 		return -ENOMEM;
 	}
 
 	tx_desc->type = ISCSI_TX_DATAOUT;
+	tx_desc->iser_header.flags = ISER_VER;
 	memcpy(&tx_desc->iscsi_header, hdr, sizeof(struct iscsi_hdr));
 
-	/* build the tx desc regd header and add it to the tx desc dto */
-	send_dto = &tx_desc->dto;
-	send_dto->task = iser_task;
-	iser_create_send_desc(iser_conn, tx_desc);
-
-	iser_reg_single(iser_conn->ib_conn->device,
-			send_dto->regd[0], DMA_TO_DEVICE);
+	/* build the tx desc */
+	iser_initialize_task_headers(task, tx_desc);
 
-	/* all data was registered for RDMA, we can use the lkey */
-	iser_dto_add_regd_buff(send_dto,
-			       &iser_task->rdma_regd[ISER_DIR_OUT],
-			       buf_offset,
-			       data_seg_len);
+	regd_buf = &iser_task->rdma_regd[ISER_DIR_OUT];
+	tx_dsg = &tx_desc->tx_sg[1];
+	tx_dsg->addr    = regd_buf->reg.va + buf_offset;
+	tx_dsg->length  = data_seg_len;
+	tx_dsg->lkey    = regd_buf->reg.lkey;
+	tx_desc->num_sge = 2;
 
 	if (buf_offset + data_seg_len > iser_task->data[ISER_DIR_OUT].data_len) {
 		iser_err("Offset:%ld & DSL:%ld in Data-Out "
@@ -485,12 +375,11 @@ int iser_send_data_out(struct iscsi_conn *conn,
 		 itt, buf_offset, data_seg_len);
 
 
-	err = iser_post_send(tx_desc);
+	err = iser_post_send(iser_conn->ib_conn, tx_desc);
 	if (!err)
 		return 0;
 
 send_data_out_error:
-	iser_dto_buffs_release(send_dto);
 	kmem_cache_free(ig.desc_cache, tx_desc);
 	iser_err("conn %p failed err %d\n",conn, err);
 	return err;
@@ -501,64 +390,44 @@ int iser_send_control(struct iscsi_conn *conn,
 {
 	struct iscsi_iser_conn *iser_conn = conn->dd_data;
 	struct iscsi_iser_task *iser_task = task->dd_data;
-	struct iser_desc *mdesc = &iser_task->desc;
-	struct iser_dto *send_dto = NULL;
+	struct iser_tx_desc *mdesc = &iser_task->desc;
 	unsigned long data_seg_len;
 	int err = 0;
-	struct iser_regd_buf *regd_buf;
 	struct iser_device *device;
-	unsigned char opcode;
-
-	if (!iser_conn_state_comp(iser_conn->ib_conn, ISER_CONN_UP)) {
-		iser_err("Failed to send, conn: 0x%p is not up\n", iser_conn->ib_conn);
-		return -EPERM;
-	}
-
-	if (iser_check_xmit(conn, task))
-		return -ENOBUFS;
 
 	/* build the tx desc regd header and add it to the tx desc dto */
 	mdesc->type = ISCSI_TX_CONTROL;
-	send_dto = &mdesc->dto;
-	send_dto->task = NULL;
-	iser_create_send_desc(iser_conn, mdesc);
+	iser_create_send_desc(iser_conn->ib_conn, mdesc);
 
 	device = iser_conn->ib_conn->device;
 
-	iser_reg_single(device, send_dto->regd[0], DMA_TO_DEVICE);
-
 	data_seg_len = ntoh24(task->hdr->dlength);
 
 	if (data_seg_len > 0) {
-		regd_buf = &mdesc->data_regd_buf;
-		memset(regd_buf, 0, sizeof(struct iser_regd_buf));
-		regd_buf->device = device;
-		regd_buf->virt_addr = task->data;
-		regd_buf->data_size = task->data_count;
-		iser_reg_single(device, regd_buf,
-				DMA_TO_DEVICE);
-		iser_dto_add_regd_buff(send_dto, regd_buf,
-				       0,
-				       data_seg_len);
+		struct ib_sge *tx_dsg = &mdesc->tx_sg[1];
+		if (task != conn->login_task) {
+			iser_err("data present on non login task!!!\n");
+			goto send_control_error;
+		}
+		memcpy(iser_conn->ib_conn->login_buf, task->data,
+							task->data_count);
+		tx_dsg->addr    = iser_conn->ib_conn->login_dma;
+		tx_dsg->length  = data_seg_len;
+		tx_dsg->lkey    = device->mr->lkey;
+		mdesc->num_sge = 2;
 	}
 
-	opcode = task->hdr->opcode & ISCSI_OPCODE_MASK;
-
-	/* post recv buffer for response if one is expected */
-	if (!(opcode == ISCSI_OP_NOOP_OUT && task->hdr->itt == RESERVED_ITT)) {
-		if (iser_post_receive_control(conn) != 0) {
-			iser_err("post_rcv_buff failed!\n");
-			err = -ENOMEM;
+	if (task == conn->login_task) {
+		err = iser_post_recvl(iser_conn->ib_conn);
+		if (err)
 			goto send_control_error;
 	}
-	}
 
-	err = iser_post_send(mdesc);
+	err = iser_post_send(iser_conn->ib_conn, mdesc);
 	if (!err)
 		return 0;
 
 send_control_error:
-	iser_dto_buffs_release(send_dto);
 	iser_err("conn %p failed err %d\n",conn, err);
 	return err;
 }
@@ -566,104 +435,71 @@ int iser_send_control(struct iscsi_conn *conn,
 /**
  * iser_rcv_dto_completion - recv DTO completion
  */
-void iser_rcv_completion(struct iser_desc *rx_desc,
-			 unsigned long dto_xfer_len)
+void iser_rcv_completion(struct iser_rx_desc *rx_desc,
+			 unsigned long rx_xfer_len,
+			 struct iser_conn *ib_conn)
 {
-	struct iser_dto *dto = &rx_desc->dto;
-	struct iscsi_iser_conn *conn = dto->ib_conn->iser_conn;
-	struct iscsi_task *task;
-	struct iscsi_iser_task *iser_task;
+	struct iscsi_iser_conn *conn = ib_conn->iser_conn;
 	struct iscsi_hdr *hdr;
-	char   *rx_data = NULL;
-	int     rx_data_len = 0;
-	unsigned char opcode;
+	u64 rx_dma;
+	int rx_buflen, outstanding, count, err;
 
-	hdr = &rx_desc->iscsi_header;
+	/* differentiate between login to all other PDUs */
+	if ((char *)rx_desc == ib_conn->login_buf) {
+		rx_dma = ib_conn->login_dma;
+		rx_buflen = ISER_RX_LOGIN_SIZE;
+	} else {
+		rx_dma = rx_desc->dma_addr;
+		rx_buflen = ISER_RX_PAYLOAD_SIZE;
+	}
 
-	iser_dbg("op 0x%x itt 0x%x\n", hdr->opcode,hdr->itt);
+	ib_dma_sync_single_for_cpu(ib_conn->device->ib_device, rx_dma,
+			rx_buflen, DMA_FROM_DEVICE);
 
-	if (dto_xfer_len > ISER_TOTAL_HEADERS_LEN) { /* we have data */
-		rx_data_len = dto_xfer_len - ISER_TOTAL_HEADERS_LEN;
-		rx_data     = dto->regd[1]->virt_addr;
-		rx_data    += dto->offset[1];
-	}
+	hdr = &rx_desc->iscsi_header;
 
-	opcode = hdr->opcode & ISCSI_OPCODE_MASK;
-
-	if (opcode == ISCSI_OP_SCSI_CMD_RSP) {
-		spin_lock(&conn->iscsi_conn->session->lock);
-		task = iscsi_itt_to_ctask(conn->iscsi_conn, hdr->itt);
-		if (task)
-			__iscsi_get_task(task);
-		spin_unlock(&conn->iscsi_conn->session->lock);
-
-		if (!task)
-			iser_err("itt can't be matched to task!!! "
-				 "conn %p opcode %d itt %d\n",
-				 conn->iscsi_conn, opcode, hdr->itt);
-		else {
-			iser_task = task->dd_data;
-			iser_dbg("itt %d task %p\n",hdr->itt, task);
-			iser_task->status = ISER_TASK_STATUS_COMPLETED;
-			iser_task_rdma_finalize(iser_task);
-			iscsi_put_task(task);
-		}
-	}
-	iser_dto_buffs_release(dto);
+	iser_dbg("op 0x%x itt 0x%x dlen %d\n", hdr->opcode,
+			hdr->itt, (int)(rx_xfer_len - ISER_HEADERS_LEN));
 
-	iscsi_iser_recv(conn->iscsi_conn, hdr, rx_data, rx_data_len);
+	iscsi_iser_recv(conn->iscsi_conn, hdr,
+		rx_desc->data, rx_xfer_len - ISER_HEADERS_LEN);
 
-	kfree(rx_desc->data);
-	kmem_cache_free(ig.desc_cache, rx_desc);
+	ib_dma_sync_single_for_device(ib_conn->device->ib_device, rx_dma,
+			rx_buflen, DMA_FROM_DEVICE);
 
 	/* decrementing conn->post_recv_buf_count only --after-- freeing the   *
 	 * task eliminates the need to worry on tasks which are completed in   *
 	 * parallel to the execution of iser_conn_term. So the code that waits *
 	 * for the posted rx bufs refcount to become zero handles everything   */
-	atomic_dec(&conn->ib_conn->post_recv_buf_count);
+	conn->ib_conn->post_recv_buf_count--;
 
-	/*
-	 * if an unexpected PDU was received then the recv wr consumed must
-	 * be replaced, this is done in the next send of a control-type PDU
-	 */
-	if (opcode == ISCSI_OP_NOOP_IN && hdr->itt == RESERVED_ITT) {
-		/* nop-in with itt = 0xffffffff */
-		atomic_inc(&conn->ib_conn->unexpected_pdu_count);
-	}
-	else if (opcode == ISCSI_OP_ASYNC_EVENT) {
-		/* asyncronous message */
-		atomic_inc(&conn->ib_conn->unexpected_pdu_count);
+	if (rx_dma == ib_conn->login_dma)
+		return;
+
+	outstanding = ib_conn->post_recv_buf_count;
+	if (outstanding + ISER_MIN_POSTED_RX <= ISER_QP_MAX_RECV_DTOS) {
+		count = min(ISER_QP_MAX_RECV_DTOS - outstanding,
+						ISER_MIN_POSTED_RX);
+		err = iser_post_recvm(ib_conn, count);
+		if (err)
+			iser_err("posting %d rx bufs err %d\n", count, err);
 	}
-	/* a reject PDU consumes the recv buf posted for the response */
 }
 
-void iser_snd_completion(struct iser_desc *tx_desc)
+void iser_snd_completion(struct iser_tx_desc *tx_desc,
+			struct iser_conn *ib_conn)
 {
-	struct iser_dto        *dto = &tx_desc->dto;
-	struct iser_conn       *ib_conn = dto->ib_conn;
-	struct iscsi_iser_conn *iser_conn = ib_conn->iser_conn;
-	struct iscsi_conn      *conn = iser_conn->iscsi_conn;
 	struct iscsi_task *task;
-	int resume_tx = 0;
-
-	iser_dbg("Initiator, Data sent dto=0x%p\n", dto);
-
-	iser_dto_buffs_release(dto);
+	struct iser_device *device = ib_conn->device;
 
-	if (tx_desc->type == ISCSI_TX_DATAOUT)
+	if (tx_desc->type == ISCSI_TX_DATAOUT) {
+		ib_dma_unmap_single(device->ib_device, tx_desc->dma_addr,
+					ISER_HEADERS_LEN, DMA_TO_DEVICE);
 		kmem_cache_free(ig.desc_cache, tx_desc);
-
-	if (atomic_read(&iser_conn->ib_conn->post_send_buf_count) ==
-	    ISER_QP_MAX_REQ_DTOS)
-		resume_tx = 1;
+	}
 
 	atomic_dec(&ib_conn->post_send_buf_count);
 
-	if (resume_tx) {
-		iser_dbg("%ld resuming tx\n",jiffies);
-		iscsi_conn_queue_work(conn);
-	}
-
 	if (tx_desc->type == ISCSI_TX_CONTROL) {
 		/* this arithmetic is legal by libiscsi dd_data allocation */
 		task = (void *) ((long)(void *)tx_desc -
@@ -692,7 +528,6 @@ void iser_task_rdma_init(struct iscsi_iser_task *iser_task)
 
 void iser_task_rdma_finalize(struct iscsi_iser_task *iser_task)
 {
-	int deferred;
 	int is_rdma_aligned = 1;
 	struct iser_regd_buf *regd;
 
@@ -710,32 +545,17 @@ void iser_task_rdma_finalize(struct iscsi_iser_task *iser_task)
 
 	if (iser_task->dir[ISER_DIR_IN]) {
 		regd = &iser_task->rdma_regd[ISER_DIR_IN];
-		deferred = iser_regd_buff_release(regd);
-		if (deferred) {
-			iser_err("%d references remain for BUF-IN rdma reg\n",
-				 atomic_read(&regd->ref_count));
-		}
+		if (regd->reg.is_fmr)
+			iser_unreg_mem(&regd->reg);
 	}
 
 	if (iser_task->dir[ISER_DIR_OUT]) {
 		regd = &iser_task->rdma_regd[ISER_DIR_OUT];
-		deferred = iser_regd_buff_release(regd);
-		if (deferred) {
-			iser_err("%d references remain for BUF-OUT rdma reg\n",
-				 atomic_read(&regd->ref_count));
-		}
+		if (regd->reg.is_fmr)
+			iser_unreg_mem(&regd->reg);
 	}
 
        /* if the data was unaligned, it was already unmapped and then copied */
        if (is_rdma_aligned)
 		iser_dma_unmap_task_data(iser_task);
 }
-
-void iser_dto_buffs_release(struct iser_dto *dto)
-{
-	int i;
-
-	for (i = 0; i < dto->regd_vector_len; i++)
-		iser_regd_buff_release(dto->regd[i]);
-}
-

drivers/infiniband/ulp/iser/iser_memory.c

@@ -40,62 +40,6 @@
 
 #define ISER_KMALLOC_THRESHOLD 0x20000 /* 128K - kmalloc limit */
 
-/**
- * Decrements the reference count for the
- * registered buffer & releases it
- *
- * returns 0 if released, 1 if deferred
- */
-int iser_regd_buff_release(struct iser_regd_buf *regd_buf)
-{
-	struct ib_device *dev;
-
-	if ((atomic_read(&regd_buf->ref_count) == 0) ||
-	    atomic_dec_and_test(&regd_buf->ref_count)) {
-		/* if we used the dma mr, unreg is just NOP */
-		if (regd_buf->reg.is_fmr)
-			iser_unreg_mem(&regd_buf->reg);
-
-		if (regd_buf->dma_addr) {
-			dev = regd_buf->device->ib_device;
-			ib_dma_unmap_single(dev,
-					 regd_buf->dma_addr,
-					 regd_buf->data_size,
-					 regd_buf->direction);
-		}
-		/* else this regd buf is associated with task which we */
-		/* dma_unmap_single/sg later */
-		return 0;
-	} else {
-		iser_dbg("Release deferred, regd.buff: 0x%p\n", regd_buf);
-		return 1;
-	}
-}
-
-/**
- * iser_reg_single - fills registered buffer descriptor with
- *		     registration information
- */
-void iser_reg_single(struct iser_device *device,
-		     struct iser_regd_buf *regd_buf,
-		     enum dma_data_direction direction)
-{
-	u64 dma_addr;
-
-	dma_addr = ib_dma_map_single(device->ib_device,
-				     regd_buf->virt_addr,
-				     regd_buf->data_size, direction);
-	BUG_ON(ib_dma_mapping_error(device->ib_device, dma_addr));
-
-	regd_buf->reg.lkey = device->mr->lkey;
-	regd_buf->reg.len  = regd_buf->data_size;
-	regd_buf->reg.va   = dma_addr;
-	regd_buf->reg.is_fmr = 0;
-
-	regd_buf->dma_addr  = dma_addr;
-	regd_buf->direction = direction;
-}
-
 /**
  * iser_start_rdma_unaligned_sg
  */
@@ -109,10 +53,10 @@ static int iser_start_rdma_unaligned_sg(struct iscsi_iser_task *iser_task,
 	unsigned long  cmd_data_len = data->data_len;
 
 	if (cmd_data_len > ISER_KMALLOC_THRESHOLD)
-		mem = (void *)__get_free_pages(GFP_NOIO,
+		mem = (void *)__get_free_pages(GFP_ATOMIC,
 		      ilog2(roundup_pow_of_two(cmd_data_len)) - PAGE_SHIFT);
 	else
-		mem = kmalloc(cmd_data_len, GFP_NOIO);
+		mem = kmalloc(cmd_data_len, GFP_ATOMIC);
 
 	if (mem == NULL) {
 		iser_err("Failed to allocate mem size %d %d for copying sglist\n",
@@ -474,9 +418,5 @@ int iser_reg_rdma_mem(struct iscsi_iser_task *iser_task,
 			return err;
 		}
 	}
-
-	/* take a reference on this regd buf such that it will not be released *
-	 * (eg in send dto completion) before we get the scsi response         */
-	atomic_inc(&regd_buf->ref_count);
 	return 0;
 }

drivers/infiniband/ulp/iser/iser_verbs.c

@@ -37,9 +37,8 @@
 #include "iscsi_iser.h"
 
 #define ISCSI_ISER_MAX_CONN	8
-#define ISER_MAX_CQ_LEN		((ISER_QP_MAX_RECV_DTOS + \
-				ISER_QP_MAX_REQ_DTOS) *   \
-				 ISCSI_ISER_MAX_CONN)
+#define ISER_MAX_RX_CQ_LEN	(ISER_QP_MAX_RECV_DTOS * ISCSI_ISER_MAX_CONN)
+#define ISER_MAX_TX_CQ_LEN	(ISER_QP_MAX_REQ_DTOS  * ISCSI_ISER_MAX_CONN)
 
 static void iser_cq_tasklet_fn(unsigned long data);
 static void iser_cq_callback(struct ib_cq *cq, void *cq_context);
@@ -67,15 +66,23 @@ static int iser_create_device_ib_res(struct iser_device *device)
 	if (IS_ERR(device->pd))
 		goto pd_err;
 
-	device->cq = ib_create_cq(device->ib_device,
+	device->rx_cq = ib_create_cq(device->ib_device,
 				  iser_cq_callback,
 				  iser_cq_event_callback,
 				  (void *)device,
-				  ISER_MAX_CQ_LEN, 0);
-	if (IS_ERR(device->cq))
-		goto cq_err;
+				  ISER_MAX_RX_CQ_LEN, 0);
+	if (IS_ERR(device->rx_cq))
+		goto rx_cq_err;
 
-	if (ib_req_notify_cq(device->cq, IB_CQ_NEXT_COMP))
+	device->tx_cq = ib_create_cq(device->ib_device,
+				  NULL, iser_cq_event_callback,
+				  (void *)device,
+				  ISER_MAX_TX_CQ_LEN, 0);
+
+	if (IS_ERR(device->tx_cq))
+		goto tx_cq_err;
+
+	if (ib_req_notify_cq(device->rx_cq, IB_CQ_NEXT_COMP))
 		goto cq_arm_err;
 
 	tasklet_init(&device->cq_tasklet,
@@ -93,8 +100,10 @@ static int iser_create_device_ib_res(struct iser_device *device)
 dma_mr_err:
 	tasklet_kill(&device->cq_tasklet);
 cq_arm_err:
-	ib_destroy_cq(device->cq);
-cq_err:
+	ib_destroy_cq(device->tx_cq);
+tx_cq_err:
+	ib_destroy_cq(device->rx_cq);
+rx_cq_err:
 	ib_dealloc_pd(device->pd);
 pd_err:
 	iser_err("failed to allocate an IB resource\n");
@@ -112,11 +121,13 @@ static void iser_free_device_ib_res(struct iser_device *device)
 	tasklet_kill(&device->cq_tasklet);
 
 	(void)ib_dereg_mr(device->mr);
-	(void)ib_destroy_cq(device->cq);
+	(void)ib_destroy_cq(device->tx_cq);
+	(void)ib_destroy_cq(device->rx_cq);
 	(void)ib_dealloc_pd(device->pd);
 
 	device->mr = NULL;
-	device->cq = NULL;
+	device->tx_cq = NULL;
+	device->rx_cq = NULL;
 	device->pd = NULL;
 }
 
@@ -129,13 +140,23 @@ static int iser_create_ib_conn_res(struct iser_conn *ib_conn)
 {
 	struct iser_device	*device;
 	struct ib_qp_init_attr	init_attr;
-	int			ret;
+	int			ret = -ENOMEM;
 	struct ib_fmr_pool_param params;
 
 	BUG_ON(ib_conn->device == NULL);
 
 	device = ib_conn->device;
 
+	ib_conn->login_buf = kmalloc(ISER_RX_LOGIN_SIZE, GFP_KERNEL);
+	if (!ib_conn->login_buf) {
+		goto alloc_err;
+		ret = -ENOMEM;
+	}
+
+	ib_conn->login_dma = ib_dma_map_single(ib_conn->device->ib_device,
+				(void *)ib_conn->login_buf, ISER_RX_LOGIN_SIZE,
+				DMA_FROM_DEVICE);
+
 	ib_conn->page_vec = kmalloc(sizeof(struct iser_page_vec) +
 				    (sizeof(u64) * (ISCSI_ISER_SG_TABLESIZE +1)),
 				    GFP_KERNEL);
@@ -169,12 +190,12 @@ static int iser_create_ib_conn_res(struct iser_conn *ib_conn)
 
 	init_attr.event_handler = iser_qp_event_callback;
 	init_attr.qp_context	= (void *)ib_conn;
-	init_attr.send_cq	= device->cq;
-	init_attr.recv_cq	= device->cq;
+	init_attr.send_cq	= device->tx_cq;
+	init_attr.recv_cq	= device->rx_cq;
 	init_attr.cap.max_send_wr  = ISER_QP_MAX_REQ_DTOS;
 	init_attr.cap.max_recv_wr  = ISER_QP_MAX_RECV_DTOS;
-	init_attr.cap.max_send_sge = MAX_REGD_BUF_VECTOR_LEN;
-	init_attr.cap.max_recv_sge = 2;
+	init_attr.cap.max_send_sge = 2;
+	init_attr.cap.max_recv_sge = 1;
 	init_attr.sq_sig_type	= IB_SIGNAL_REQ_WR;
 	init_attr.qp_type	= IB_QPT_RC;
 
@@ -192,6 +213,7 @@ static int iser_create_ib_conn_res(struct iser_conn *ib_conn)
 	(void)ib_destroy_fmr_pool(ib_conn->fmr_pool);
 fmr_pool_err:
 	kfree(ib_conn->page_vec);
+	kfree(ib_conn->login_buf);
 alloc_err:
 	iser_err("unable to alloc mem or create resource, err %d\n", ret);
 	return ret;
@@ -278,17 +300,6 @@ static void iser_device_try_release(struct iser_device *device)
 	mutex_unlock(&ig.device_list_mutex);
 }
 
-int iser_conn_state_comp(struct iser_conn *ib_conn,
-			enum iser_ib_conn_state comp)
-{
-	int ret;
-
-	spin_lock_bh(&ib_conn->lock);
-	ret = (ib_conn->state == comp);
-	spin_unlock_bh(&ib_conn->lock);
-	return ret;
-}
-
 static int iser_conn_state_comp_exch(struct iser_conn *ib_conn,
 				     enum iser_ib_conn_state comp,
 				     enum iser_ib_conn_state exch)
@@ -314,7 +325,7 @@ static void iser_conn_release(struct iser_conn *ib_conn)
 	mutex_lock(&ig.connlist_mutex);
 	list_del(&ib_conn->conn_list);
 	mutex_unlock(&ig.connlist_mutex);
-
+	iser_free_rx_descriptors(ib_conn);
 	iser_free_ib_conn_res(ib_conn);
 	ib_conn->device = NULL;
 	/* on EVENT_ADDR_ERROR there's no device yet for this conn */
@@ -442,7 +453,7 @@ static void iser_disconnected_handler(struct rdma_cm_id *cma_id)
 				   ISCSI_ERR_CONN_FAILED);
 
 	/* Complete the termination process if no posts are pending */
-	if ((atomic_read(&ib_conn->post_recv_buf_count) == 0) &&
+	if (ib_conn->post_recv_buf_count == 0 &&
 	    (atomic_read(&ib_conn->post_send_buf_count) == 0)) {
 		ib_conn->state = ISER_CONN_DOWN;
 		wake_up_interruptible(&ib_conn->wait);
@@ -489,9 +500,8 @@ void iser_conn_init(struct iser_conn *ib_conn)
 {
 	ib_conn->state = ISER_CONN_INIT;
 	init_waitqueue_head(&ib_conn->wait);
-	atomic_set(&ib_conn->post_recv_buf_count, 0);
+	ib_conn->post_recv_buf_count = 0;
 	atomic_set(&ib_conn->post_send_buf_count, 0);
-	atomic_set(&ib_conn->unexpected_pdu_count, 0);
 	atomic_set(&ib_conn->refcount, 1);
 	INIT_LIST_HEAD(&ib_conn->conn_list);
 	spin_lock_init(&ib_conn->lock);
@@ -626,136 +636,97 @@ void iser_unreg_mem(struct iser_mem_reg *reg)
 	reg->mem_h = NULL;
 }
 
-/**
- * iser_dto_to_iov - builds IOV from a dto descriptor
- */
-static void iser_dto_to_iov(struct iser_dto *dto, struct ib_sge *iov, int iov_len)
+int iser_post_recvl(struct iser_conn *ib_conn)
 {
-	int		     i;
-	struct ib_sge	     *sge;
-	struct iser_regd_buf *regd_buf;
-
-	if (dto->regd_vector_len > iov_len) {
-		iser_err("iov size %d too small for posting dto of len %d\n",
-			 iov_len, dto->regd_vector_len);
-		BUG();
-	}
+	struct ib_recv_wr rx_wr, *rx_wr_failed;
+	struct ib_sge	  sge;
+	int ib_ret;
 
-	for (i = 0; i < dto->regd_vector_len; i++) {
-		sge	    = &iov[i];
-		regd_buf  = dto->regd[i];
-
-		sge->addr   = regd_buf->reg.va;
-		sge->length = regd_buf->reg.len;
-		sge->lkey   = regd_buf->reg.lkey;
-
-		if (dto->used_sz[i] > 0)  /* Adjust size */
-			sge->length = dto->used_sz[i];
-
-		/* offset and length should not exceed the regd buf length */
-		if (sge->length + dto->offset[i] > regd_buf->reg.len) {
-			iser_err("Used len:%ld + offset:%d, exceed reg.buf.len:"
-				 "%ld in dto:0x%p [%d], va:0x%08lX\n",
-				 (unsigned long)sge->length, dto->offset[i],
-				 (unsigned long)regd_buf->reg.len, dto, i,
-				 (unsigned long)sge->addr);
-			BUG();
-		}
+	sge.addr   = ib_conn->login_dma;
+	sge.length = ISER_RX_LOGIN_SIZE;
+	sge.lkey   = ib_conn->device->mr->lkey;
+
+	rx_wr.wr_id   = (unsigned long)ib_conn->login_buf;
+	rx_wr.sg_list = &sge;
+	rx_wr.num_sge = 1;
+	rx_wr.next    = NULL;
 
-		sge->addr += dto->offset[i]; /* Adjust offset */
+	ib_conn->post_recv_buf_count++;
+	ib_ret	= ib_post_recv(ib_conn->qp, &rx_wr, &rx_wr_failed);
+	if (ib_ret) {
+		iser_err("ib_post_recv failed ret=%d\n", ib_ret);
+		ib_conn->post_recv_buf_count--;
 	}
+	return ib_ret;
 }
 
-/**
- * iser_post_recv - Posts a receive buffer.
- *
- * returns 0 on success, -1 on failure
- */
-int iser_post_recv(struct iser_desc *rx_desc)
+int iser_post_recvm(struct iser_conn *ib_conn, int count)
 {
-	int		  ib_ret, ret_val = 0;
-	struct ib_recv_wr recv_wr, *recv_wr_failed;
-	struct ib_sge	  iov[2];
-	struct iser_conn  *ib_conn;
-	struct iser_dto   *recv_dto = &rx_desc->dto;
-
-	/* Retrieve conn */
-	ib_conn = recv_dto->ib_conn;
-
-	iser_dto_to_iov(recv_dto, iov, 2);
+	struct ib_recv_wr *rx_wr, *rx_wr_failed;
+	int i, ib_ret;
+	unsigned int my_rx_head = ib_conn->rx_desc_head;
+	struct iser_rx_desc *rx_desc;
+
+	for (rx_wr = ib_conn->rx_wr, i = 0; i < count; i++, rx_wr++) {
+		rx_desc		= &ib_conn->rx_descs[my_rx_head];
+		rx_wr->wr_id	= (unsigned long)rx_desc;
+		rx_wr->sg_list	= &rx_desc->rx_sg;
+		rx_wr->num_sge	= 1;
+		rx_wr->next	= rx_wr + 1;
+		my_rx_head = (my_rx_head + 1) & (ISER_QP_MAX_RECV_DTOS - 1);
+	}
 
-	recv_wr.next	= NULL;
-	recv_wr.sg_list = iov;
-	recv_wr.num_sge = recv_dto->regd_vector_len;
-	recv_wr.wr_id	= (unsigned long)rx_desc;
+	rx_wr--;
+	rx_wr->next = NULL; /* mark end of work requests list */
 
-	atomic_inc(&ib_conn->post_recv_buf_count);
-	ib_ret	= ib_post_recv(ib_conn->qp, &recv_wr, &recv_wr_failed);
+	ib_conn->post_recv_buf_count += count;
+	ib_ret	= ib_post_recv(ib_conn->qp, ib_conn->rx_wr, &rx_wr_failed);
 	if (ib_ret) {
 		iser_err("ib_post_recv failed ret=%d\n", ib_ret);
-		atomic_dec(&ib_conn->post_recv_buf_count);
-		ret_val = -1;
-	}
-
-	return ret_val;
+		ib_conn->post_recv_buf_count -= count;
+	} else
+		ib_conn->rx_desc_head = my_rx_head;
+	return ib_ret;
 }
 
+
 /**
  * iser_start_send - Initiate a Send DTO operation
  *
  * returns 0 on success, -1 on failure
  */
-int iser_post_send(struct iser_desc *tx_desc)
+int iser_post_send(struct iser_conn *ib_conn, struct iser_tx_desc *tx_desc)
 {
-	int		  ib_ret, ret_val = 0;
+	int		  ib_ret;
 	struct ib_send_wr send_wr, *send_wr_failed;
-	struct ib_sge	  iov[MAX_REGD_BUF_VECTOR_LEN];
-	struct iser_conn  *ib_conn;
-	struct iser_dto   *dto = &tx_desc->dto;
-
-	ib_conn = dto->ib_conn;
 
-	iser_dto_to_iov(dto, iov, MAX_REGD_BUF_VECTOR_LEN);
+	ib_dma_sync_single_for_device(ib_conn->device->ib_device,
+		tx_desc->dma_addr, ISER_HEADERS_LEN, DMA_TO_DEVICE);
 
 	send_wr.next	   = NULL;
 	send_wr.wr_id	   = (unsigned long)tx_desc;
-	send_wr.sg_list	   = iov;
-	send_wr.num_sge	   = dto->regd_vector_len;
+	send_wr.sg_list	   = tx_desc->tx_sg;
+	send_wr.num_sge	   = tx_desc->num_sge;
 	send_wr.opcode	   = IB_WR_SEND;
-	send_wr.send_flags = dto->notify_enable ? IB_SEND_SIGNALED : 0;
+	send_wr.send_flags = IB_SEND_SIGNALED;
 
 	atomic_inc(&ib_conn->post_send_buf_count);
 
 	ib_ret = ib_post_send(ib_conn->qp, &send_wr, &send_wr_failed);
 	if (ib_ret) {
-		iser_err("Failed to start SEND DTO, dto: 0x%p, IOV len: %d\n",
-			 dto, dto->regd_vector_len);
 		iser_err("ib_post_send failed, ret:%d\n", ib_ret);
 		atomic_dec(&ib_conn->post_send_buf_count);
-		ret_val = -1;
 	}
-
-	return ret_val;
+	return ib_ret;
 }
 
-static void iser_handle_comp_error(struct iser_desc *desc)
+static void iser_handle_comp_error(struct iser_tx_desc *desc,
+				struct iser_conn *ib_conn)
 {
-	struct iser_dto  *dto     = &desc->dto;
-	struct iser_conn *ib_conn = dto->ib_conn;
-
-	iser_dto_buffs_release(dto);
-
-	if (desc->type == ISCSI_RX) {
-		kfree(desc->data);
-		kmem_cache_free(ig.desc_cache, desc);
-		atomic_dec(&ib_conn->post_recv_buf_count);
-	} else { /* type is TX control/command/dataout */
-		if (desc->type == ISCSI_TX_DATAOUT)
+	if (desc && desc->type == ISCSI_TX_DATAOUT)
 		kmem_cache_free(ig.desc_cache, desc);
-		atomic_dec(&ib_conn->post_send_buf_count);
-	}
 
-	if (atomic_read(&ib_conn->post_recv_buf_count) == 0 &&
+	if (ib_conn->post_recv_buf_count == 0 &&
 	    atomic_read(&ib_conn->post_send_buf_count) == 0) {
 		/* getting here when the state is UP means that the conn is *
 		 * being terminated asynchronously from the iSCSI layer's   *
@@ -774,32 +745,74 @@ static void iser_handle_comp_error(struct iser_desc *desc)
 	}
 }
 
+static int iser_drain_tx_cq(struct iser_device  *device)
+{
+	struct ib_cq  *cq = device->tx_cq;
+	struct ib_wc  wc;
+	struct iser_tx_desc *tx_desc;
+	struct iser_conn *ib_conn;
+	int completed_tx = 0;
+
+	while (ib_poll_cq(cq, 1, &wc) == 1) {
+		tx_desc	= (struct iser_tx_desc *) (unsigned long) wc.wr_id;
+		ib_conn = wc.qp->qp_context;
+		if (wc.status == IB_WC_SUCCESS) {
+			if (wc.opcode == IB_WC_SEND)
+				iser_snd_completion(tx_desc, ib_conn);
+			else
+				iser_err("expected opcode %d got %d\n",
+					IB_WC_SEND, wc.opcode);
+		} else {
+			iser_err("tx id %llx status %d vend_err %x\n",
+				wc.wr_id, wc.status, wc.vendor_err);
+			atomic_dec(&ib_conn->post_send_buf_count);
+			iser_handle_comp_error(tx_desc, ib_conn);
+		}
+		completed_tx++;
+	}
+	return completed_tx;
+}
+
+
 static void iser_cq_tasklet_fn(unsigned long data)
 {
 	 struct iser_device  *device = (struct iser_device *)data;
-	 struct ib_cq	     *cq = device->cq;
+	 struct ib_cq	     *cq = device->rx_cq;
 	 struct ib_wc	     wc;
-	 struct iser_desc    *desc;
+	 struct iser_rx_desc *desc;
 	 unsigned long	     xfer_len;
+	struct iser_conn *ib_conn;
+	int completed_tx, completed_rx;
+	completed_tx = completed_rx = 0;
 
 	while (ib_poll_cq(cq, 1, &wc) == 1) {
-		desc	 = (struct iser_desc *) (unsigned long) wc.wr_id;
+		desc	 = (struct iser_rx_desc *) (unsigned long) wc.wr_id;
 		BUG_ON(desc == NULL);
-
+		ib_conn = wc.qp->qp_context;
 		if (wc.status == IB_WC_SUCCESS) {
-			if (desc->type == ISCSI_RX) {
+			if (wc.opcode == IB_WC_RECV) {
 				xfer_len = (unsigned long)wc.byte_len;
-				iser_rcv_completion(desc, xfer_len);
-			} else /* type == ISCSI_TX_CONTROL/SCSI_CMD/DOUT */
-				iser_snd_completion(desc);
+				iser_rcv_completion(desc, xfer_len, ib_conn);
+			} else
+				iser_err("expected opcode %d got %d\n",
+					IB_WC_RECV, wc.opcode);
 		} else {
-			iser_err("comp w. error op %d status %d\n",desc->type,wc.status);
-			iser_handle_comp_error(desc);
+			if (wc.status != IB_WC_WR_FLUSH_ERR)
+				iser_err("rx id %llx status %d vend_err %x\n",
+					wc.wr_id, wc.status, wc.vendor_err);
+			ib_conn->post_recv_buf_count--;
+			iser_handle_comp_error(NULL, ib_conn);
 		}
+		completed_rx++;
+		if (!(completed_rx & 63))
+			completed_tx += iser_drain_tx_cq(device);
 	}
 	/* #warning "it is assumed here that arming CQ only once its empty" *
 	 * " would not cause interrupts to be missed"                       */
 	ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
+
+	completed_tx += iser_drain_tx_cq(device);
+	iser_dbg("got %d rx %d tx completions\n", completed_rx, completed_tx);
 }
 
 static void iser_cq_callback(struct ib_cq *cq, void *cq_context)