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_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;
 }