Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / scsi / bnx2i / bnx2i_iscsi.c
diff --git a/kernel/drivers/scsi/bnx2i/bnx2i_iscsi.c b/kernel/drivers/scsi/bnx2i/bnx2i_iscsi.c
new file mode 100644 (file)
index 0000000..e53078d
--- /dev/null
@@ -0,0 +1,2305 @@
+/*
+ * bnx2i_iscsi.c: QLogic NetXtreme II iSCSI driver.
+ *
+ * Copyright (c) 2006 - 2013 Broadcom Corporation
+ * Copyright (c) 2007, 2008 Red Hat, Inc.  All rights reserved.
+ * Copyright (c) 2007, 2008 Mike Christie
+ * Copyright (c) 2014, QLogic Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation.
+ *
+ * Written by: Anil Veerabhadrappa (anilgv@broadcom.com)
+ * Previously Maintained by: Eddie Wai (eddie.wai@broadcom.com)
+ * Maintained by: QLogic-Storage-Upstream@qlogic.com
+ */
+
+#include <linux/slab.h>
+#include <scsi/scsi_tcq.h>
+#include <scsi/libiscsi.h>
+#include "bnx2i.h"
+
+struct scsi_transport_template *bnx2i_scsi_xport_template;
+struct iscsi_transport bnx2i_iscsi_transport;
+static struct scsi_host_template bnx2i_host_template;
+
+/*
+ * Global endpoint resource info
+ */
+static DEFINE_SPINLOCK(bnx2i_resc_lock); /* protects global resources */
+
+DECLARE_PER_CPU(struct bnx2i_percpu_s, bnx2i_percpu);
+
+static int bnx2i_adapter_ready(struct bnx2i_hba *hba)
+{
+       int retval = 0;
+
+       if (!hba || !test_bit(ADAPTER_STATE_UP, &hba->adapter_state) ||
+           test_bit(ADAPTER_STATE_GOING_DOWN, &hba->adapter_state) ||
+           test_bit(ADAPTER_STATE_LINK_DOWN, &hba->adapter_state))
+               retval = -EPERM;
+       return retval;
+}
+
+/**
+ * bnx2i_get_write_cmd_bd_idx - identifies various BD bookmarks
+ * @cmd:               iscsi cmd struct pointer
+ * @buf_off:           absolute buffer offset
+ * @start_bd_off:      u32 pointer to return the offset within the BD
+ *                     indicated by 'start_bd_idx' on which 'buf_off' falls
+ * @start_bd_idx:      index of the BD on which 'buf_off' falls
+ *
+ * identifies & marks various bd info for scsi command's imm data,
+ * unsolicited data and the first solicited data seq.
+ */
+static void bnx2i_get_write_cmd_bd_idx(struct bnx2i_cmd *cmd, u32 buf_off,
+                                      u32 *start_bd_off, u32 *start_bd_idx)
+{
+       struct iscsi_bd *bd_tbl = cmd->io_tbl.bd_tbl;
+       u32 cur_offset = 0;
+       u32 cur_bd_idx = 0;
+
+       if (buf_off) {
+               while (buf_off >= (cur_offset + bd_tbl->buffer_length)) {
+                       cur_offset += bd_tbl->buffer_length;
+                       cur_bd_idx++;
+                       bd_tbl++;
+               }
+       }
+
+       *start_bd_off = buf_off - cur_offset;
+       *start_bd_idx = cur_bd_idx;
+}
+
+/**
+ * bnx2i_setup_write_cmd_bd_info - sets up BD various information
+ * @task:      transport layer's cmd struct pointer
+ *
+ * identifies & marks various bd info for scsi command's immediate data,
+ * unsolicited data and first solicited data seq which includes BD start
+ * index & BD buf off. his function takes into account iscsi parameter such
+ * as immediate data and unsolicited data is support on this connection.
+ */
+static void bnx2i_setup_write_cmd_bd_info(struct iscsi_task *task)
+{
+       struct bnx2i_cmd *cmd = task->dd_data;
+       u32 start_bd_offset;
+       u32 start_bd_idx;
+       u32 buffer_offset = 0;
+       u32 cmd_len = cmd->req.total_data_transfer_length;
+
+       /* if ImmediateData is turned off & IntialR2T is turned on,
+        * there will be no immediate or unsolicited data, just return.
+        */
+       if (!iscsi_task_has_unsol_data(task) && !task->imm_count)
+               return;
+
+       /* Immediate data */
+       buffer_offset += task->imm_count;
+       if (task->imm_count == cmd_len)
+               return;
+
+       if (iscsi_task_has_unsol_data(task)) {
+               bnx2i_get_write_cmd_bd_idx(cmd, buffer_offset,
+                                          &start_bd_offset, &start_bd_idx);
+               cmd->req.ud_buffer_offset = start_bd_offset;
+               cmd->req.ud_start_bd_index = start_bd_idx;
+               buffer_offset += task->unsol_r2t.data_length;
+       }
+
+       if (buffer_offset != cmd_len) {
+               bnx2i_get_write_cmd_bd_idx(cmd, buffer_offset,
+                                          &start_bd_offset, &start_bd_idx);
+               if ((start_bd_offset > task->conn->session->first_burst) ||
+                   (start_bd_idx > scsi_sg_count(cmd->scsi_cmd))) {
+                       int i = 0;
+
+                       iscsi_conn_printk(KERN_ALERT, task->conn,
+                                         "bnx2i- error, buf offset 0x%x "
+                                         "bd_valid %d use_sg %d\n",
+                                         buffer_offset, cmd->io_tbl.bd_valid,
+                                         scsi_sg_count(cmd->scsi_cmd));
+                       for (i = 0; i < cmd->io_tbl.bd_valid; i++)
+                               iscsi_conn_printk(KERN_ALERT, task->conn,
+                                                 "bnx2i err, bd[%d]: len %x\n",
+                                                 i, cmd->io_tbl.bd_tbl[i].\
+                                                 buffer_length);
+               }
+               cmd->req.sd_buffer_offset = start_bd_offset;
+               cmd->req.sd_start_bd_index = start_bd_idx;
+       }
+}
+
+
+
+/**
+ * bnx2i_map_scsi_sg - maps IO buffer and prepares the BD table
+ * @hba:       adapter instance
+ * @cmd:       iscsi cmd struct pointer
+ *
+ * map SG list
+ */
+static int bnx2i_map_scsi_sg(struct bnx2i_hba *hba, struct bnx2i_cmd *cmd)
+{
+       struct scsi_cmnd *sc = cmd->scsi_cmd;
+       struct iscsi_bd *bd = cmd->io_tbl.bd_tbl;
+       struct scatterlist *sg;
+       int byte_count = 0;
+       int bd_count = 0;
+       int sg_count;
+       int sg_len;
+       u64 addr;
+       int i;
+
+       BUG_ON(scsi_sg_count(sc) > ISCSI_MAX_BDS_PER_CMD);
+
+       sg_count = scsi_dma_map(sc);
+
+       scsi_for_each_sg(sc, sg, sg_count, i) {
+               sg_len = sg_dma_len(sg);
+               addr = (u64) sg_dma_address(sg);
+               bd[bd_count].buffer_addr_lo = addr & 0xffffffff;
+               bd[bd_count].buffer_addr_hi = addr >> 32;
+               bd[bd_count].buffer_length = sg_len;
+               bd[bd_count].flags = 0;
+               if (bd_count == 0)
+                       bd[bd_count].flags = ISCSI_BD_FIRST_IN_BD_CHAIN;
+
+               byte_count += sg_len;
+               bd_count++;
+       }
+
+       if (bd_count)
+               bd[bd_count - 1].flags |= ISCSI_BD_LAST_IN_BD_CHAIN;
+
+       BUG_ON(byte_count != scsi_bufflen(sc));
+       return bd_count;
+}
+
+/**
+ * bnx2i_iscsi_map_sg_list - maps SG list
+ * @cmd:       iscsi cmd struct pointer
+ *
+ * creates BD list table for the command
+ */
+static void bnx2i_iscsi_map_sg_list(struct bnx2i_cmd *cmd)
+{
+       int bd_count;
+
+       bd_count  = bnx2i_map_scsi_sg(cmd->conn->hba, cmd);
+       if (!bd_count) {
+               struct iscsi_bd *bd = cmd->io_tbl.bd_tbl;
+
+               bd[0].buffer_addr_lo = bd[0].buffer_addr_hi = 0;
+               bd[0].buffer_length = bd[0].flags = 0;
+       }
+       cmd->io_tbl.bd_valid = bd_count;
+}
+
+
+/**
+ * bnx2i_iscsi_unmap_sg_list - unmaps SG list
+ * @cmd:       iscsi cmd struct pointer
+ *
+ * unmap IO buffers and invalidate the BD table
+ */
+void bnx2i_iscsi_unmap_sg_list(struct bnx2i_cmd *cmd)
+{
+       struct scsi_cmnd *sc = cmd->scsi_cmd;
+
+       if (cmd->io_tbl.bd_valid && sc) {
+               scsi_dma_unmap(sc);
+               cmd->io_tbl.bd_valid = 0;
+       }
+}
+
+static void bnx2i_setup_cmd_wqe_template(struct bnx2i_cmd *cmd)
+{
+       memset(&cmd->req, 0x00, sizeof(cmd->req));
+       cmd->req.op_code = 0xFF;
+       cmd->req.bd_list_addr_lo = (u32) cmd->io_tbl.bd_tbl_dma;
+       cmd->req.bd_list_addr_hi =
+               (u32) ((u64) cmd->io_tbl.bd_tbl_dma >> 32);
+
+}
+
+
+/**
+ * bnx2i_bind_conn_to_iscsi_cid - bind conn structure to 'iscsi_cid'
+ * @hba:       pointer to adapter instance
+ * @conn:      pointer to iscsi connection
+ * @iscsi_cid: iscsi context ID, range 0 - (MAX_CONN - 1)
+ *
+ * update iscsi cid table entry with connection pointer. This enables
+ *     driver to quickly get hold of connection structure pointer in
+ *     completion/interrupt thread using iscsi context ID
+ */
+static int bnx2i_bind_conn_to_iscsi_cid(struct bnx2i_hba *hba,
+                                       struct bnx2i_conn *bnx2i_conn,
+                                       u32 iscsi_cid)
+{
+       if (hba && hba->cid_que.conn_cid_tbl[iscsi_cid]) {
+               iscsi_conn_printk(KERN_ALERT, bnx2i_conn->cls_conn->dd_data,
+                                "conn bind - entry #%d not free\n", iscsi_cid);
+               return -EBUSY;
+       }
+
+       hba->cid_que.conn_cid_tbl[iscsi_cid] = bnx2i_conn;
+       return 0;
+}
+
+
+/**
+ * bnx2i_get_conn_from_id - maps an iscsi cid to corresponding conn ptr
+ * @hba:       pointer to adapter instance
+ * @iscsi_cid: iscsi context ID, range 0 - (MAX_CONN - 1)
+ */
+struct bnx2i_conn *bnx2i_get_conn_from_id(struct bnx2i_hba *hba,
+                                         u16 iscsi_cid)
+{
+       if (!hba->cid_que.conn_cid_tbl) {
+               printk(KERN_ERR "bnx2i: ERROR - missing conn<->cid table\n");
+               return NULL;
+
+       } else if (iscsi_cid >= hba->max_active_conns) {
+               printk(KERN_ERR "bnx2i: wrong cid #%d\n", iscsi_cid);
+               return NULL;
+       }
+       return hba->cid_que.conn_cid_tbl[iscsi_cid];
+}
+
+
+/**
+ * bnx2i_alloc_iscsi_cid - allocates a iscsi_cid from free pool
+ * @hba:       pointer to adapter instance
+ */
+static u32 bnx2i_alloc_iscsi_cid(struct bnx2i_hba *hba)
+{
+       int idx;
+
+       if (!hba->cid_que.cid_free_cnt)
+               return -1;
+
+       idx = hba->cid_que.cid_q_cons_idx;
+       hba->cid_que.cid_q_cons_idx++;
+       if (hba->cid_que.cid_q_cons_idx == hba->cid_que.cid_q_max_idx)
+               hba->cid_que.cid_q_cons_idx = 0;
+
+       hba->cid_que.cid_free_cnt--;
+       return hba->cid_que.cid_que[idx];
+}
+
+
+/**
+ * bnx2i_free_iscsi_cid - returns tcp port to free list
+ * @hba:               pointer to adapter instance
+ * @iscsi_cid:         iscsi context ID to free
+ */
+static void bnx2i_free_iscsi_cid(struct bnx2i_hba *hba, u16 iscsi_cid)
+{
+       int idx;
+
+       if (iscsi_cid == (u16) -1)
+               return;
+
+       hba->cid_que.cid_free_cnt++;
+
+       idx = hba->cid_que.cid_q_prod_idx;
+       hba->cid_que.cid_que[idx] = iscsi_cid;
+       hba->cid_que.conn_cid_tbl[iscsi_cid] = NULL;
+       hba->cid_que.cid_q_prod_idx++;
+       if (hba->cid_que.cid_q_prod_idx == hba->cid_que.cid_q_max_idx)
+               hba->cid_que.cid_q_prod_idx = 0;
+}
+
+
+/**
+ * bnx2i_setup_free_cid_que - sets up free iscsi cid queue
+ * @hba:       pointer to adapter instance
+ *
+ * allocates memory for iscsi cid queue & 'cid - conn ptr' mapping table,
+ *     and initialize table attributes
+ */
+static int bnx2i_setup_free_cid_que(struct bnx2i_hba *hba)
+{
+       int mem_size;
+       int i;
+
+       mem_size = hba->max_active_conns * sizeof(u32);
+       mem_size = (mem_size + (PAGE_SIZE - 1)) & PAGE_MASK;
+
+       hba->cid_que.cid_que_base = kmalloc(mem_size, GFP_KERNEL);
+       if (!hba->cid_que.cid_que_base)
+               return -ENOMEM;
+
+       mem_size = hba->max_active_conns * sizeof(struct bnx2i_conn *);
+       mem_size = (mem_size + (PAGE_SIZE - 1)) & PAGE_MASK;
+       hba->cid_que.conn_cid_tbl = kmalloc(mem_size, GFP_KERNEL);
+       if (!hba->cid_que.conn_cid_tbl) {
+               kfree(hba->cid_que.cid_que_base);
+               hba->cid_que.cid_que_base = NULL;
+               return -ENOMEM;
+       }
+
+       hba->cid_que.cid_que = (u32 *)hba->cid_que.cid_que_base;
+       hba->cid_que.cid_q_prod_idx = 0;
+       hba->cid_que.cid_q_cons_idx = 0;
+       hba->cid_que.cid_q_max_idx = hba->max_active_conns;
+       hba->cid_que.cid_free_cnt = hba->max_active_conns;
+
+       for (i = 0; i < hba->max_active_conns; i++) {
+               hba->cid_que.cid_que[i] = i;
+               hba->cid_que.conn_cid_tbl[i] = NULL;
+       }
+       return 0;
+}
+
+
+/**
+ * bnx2i_release_free_cid_que - releases 'iscsi_cid' queue resources
+ * @hba:       pointer to adapter instance
+ */
+static void bnx2i_release_free_cid_que(struct bnx2i_hba *hba)
+{
+       kfree(hba->cid_que.cid_que_base);
+       hba->cid_que.cid_que_base = NULL;
+
+       kfree(hba->cid_que.conn_cid_tbl);
+       hba->cid_que.conn_cid_tbl = NULL;
+}
+
+
+/**
+ * bnx2i_alloc_ep - allocates ep structure from global pool
+ * @hba:       pointer to adapter instance
+ *
+ * routine allocates a free endpoint structure from global pool and
+ *     a tcp port to be used for this connection.  Global resource lock,
+ *     'bnx2i_resc_lock' is held while accessing shared global data structures
+ */
+static struct iscsi_endpoint *bnx2i_alloc_ep(struct bnx2i_hba *hba)
+{
+       struct iscsi_endpoint *ep;
+       struct bnx2i_endpoint *bnx2i_ep;
+       u32 ec_div;
+
+       ep = iscsi_create_endpoint(sizeof(*bnx2i_ep));
+       if (!ep) {
+               printk(KERN_ERR "bnx2i: Could not allocate ep\n");
+               return NULL;
+       }
+
+       bnx2i_ep = ep->dd_data;
+       bnx2i_ep->cls_ep = ep;
+       INIT_LIST_HEAD(&bnx2i_ep->link);
+       bnx2i_ep->state = EP_STATE_IDLE;
+       bnx2i_ep->ep_iscsi_cid = (u16) -1;
+       bnx2i_ep->hba = hba;
+       bnx2i_ep->hba_age = hba->age;
+
+       ec_div = event_coal_div;
+       while (ec_div >>= 1)
+               bnx2i_ep->ec_shift += 1;
+
+       hba->ofld_conns_active++;
+       init_waitqueue_head(&bnx2i_ep->ofld_wait);
+       return ep;
+}
+
+
+/**
+ * bnx2i_free_ep - free endpoint
+ * @ep:                pointer to iscsi endpoint structure
+ */
+static void bnx2i_free_ep(struct iscsi_endpoint *ep)
+{
+       struct bnx2i_endpoint *bnx2i_ep = ep->dd_data;
+       unsigned long flags;
+
+       spin_lock_irqsave(&bnx2i_resc_lock, flags);
+       bnx2i_ep->state = EP_STATE_IDLE;
+       bnx2i_ep->hba->ofld_conns_active--;
+
+       if (bnx2i_ep->ep_iscsi_cid != (u16) -1)
+               bnx2i_free_iscsi_cid(bnx2i_ep->hba, bnx2i_ep->ep_iscsi_cid);
+
+       if (bnx2i_ep->conn) {
+               bnx2i_ep->conn->ep = NULL;
+               bnx2i_ep->conn = NULL;
+       }
+
+       bnx2i_ep->hba = NULL;
+       spin_unlock_irqrestore(&bnx2i_resc_lock, flags);
+       iscsi_destroy_endpoint(ep);
+}
+
+
+/**
+ * bnx2i_alloc_bdt - allocates buffer descriptor (BD) table for the command
+ * @hba:       adapter instance pointer
+ * @session:   iscsi session pointer
+ * @cmd:       iscsi command structure
+ */
+static int bnx2i_alloc_bdt(struct bnx2i_hba *hba, struct iscsi_session *session,
+                          struct bnx2i_cmd *cmd)
+{
+       struct io_bdt *io = &cmd->io_tbl;
+       struct iscsi_bd *bd;
+
+       io->bd_tbl = dma_alloc_coherent(&hba->pcidev->dev,
+                                       ISCSI_MAX_BDS_PER_CMD * sizeof(*bd),
+                                       &io->bd_tbl_dma, GFP_KERNEL);
+       if (!io->bd_tbl) {
+               iscsi_session_printk(KERN_ERR, session, "Could not "
+                                    "allocate bdt.\n");
+               return -ENOMEM;
+       }
+       io->bd_valid = 0;
+       return 0;
+}
+
+/**
+ * bnx2i_destroy_cmd_pool - destroys iscsi command pool and release BD table
+ * @hba:       adapter instance pointer
+ * @session:   iscsi session pointer
+ * @cmd:       iscsi command structure
+ */
+static void bnx2i_destroy_cmd_pool(struct bnx2i_hba *hba,
+                                  struct iscsi_session *session)
+{
+       int i;
+
+       for (i = 0; i < session->cmds_max; i++) {
+               struct iscsi_task *task = session->cmds[i];
+               struct bnx2i_cmd *cmd = task->dd_data;
+
+               if (cmd->io_tbl.bd_tbl)
+                       dma_free_coherent(&hba->pcidev->dev,
+                                         ISCSI_MAX_BDS_PER_CMD *
+                                         sizeof(struct iscsi_bd),
+                                         cmd->io_tbl.bd_tbl,
+                                         cmd->io_tbl.bd_tbl_dma);
+       }
+
+}
+
+
+/**
+ * bnx2i_setup_cmd_pool - sets up iscsi command pool for the session
+ * @hba:       adapter instance pointer
+ * @session:   iscsi session pointer
+ */
+static int bnx2i_setup_cmd_pool(struct bnx2i_hba *hba,
+                               struct iscsi_session *session)
+{
+       int i;
+
+       for (i = 0; i < session->cmds_max; i++) {
+               struct iscsi_task *task = session->cmds[i];
+               struct bnx2i_cmd *cmd = task->dd_data;
+
+               task->hdr = &cmd->hdr;
+               task->hdr_max = sizeof(struct iscsi_hdr);
+
+               if (bnx2i_alloc_bdt(hba, session, cmd))
+                       goto free_bdts;
+       }
+
+       return 0;
+
+free_bdts:
+       bnx2i_destroy_cmd_pool(hba, session);
+       return -ENOMEM;
+}
+
+
+/**
+ * bnx2i_setup_mp_bdt - allocate BD table resources
+ * @hba:       pointer to adapter structure
+ *
+ * Allocate memory for dummy buffer and associated BD
+ * table to be used by middle path (MP) requests
+ */
+static int bnx2i_setup_mp_bdt(struct bnx2i_hba *hba)
+{
+       int rc = 0;
+       struct iscsi_bd *mp_bdt;
+       u64 addr;
+
+       hba->mp_bd_tbl = dma_alloc_coherent(&hba->pcidev->dev, CNIC_PAGE_SIZE,
+                                           &hba->mp_bd_dma, GFP_KERNEL);
+       if (!hba->mp_bd_tbl) {
+               printk(KERN_ERR "unable to allocate Middle Path BDT\n");
+               rc = -1;
+               goto out;
+       }
+
+       hba->dummy_buffer = dma_alloc_coherent(&hba->pcidev->dev,
+                                              CNIC_PAGE_SIZE,
+                                              &hba->dummy_buf_dma, GFP_KERNEL);
+       if (!hba->dummy_buffer) {
+               printk(KERN_ERR "unable to alloc Middle Path Dummy Buffer\n");
+               dma_free_coherent(&hba->pcidev->dev, CNIC_PAGE_SIZE,
+                                 hba->mp_bd_tbl, hba->mp_bd_dma);
+               hba->mp_bd_tbl = NULL;
+               rc = -1;
+               goto out;
+       }
+
+       mp_bdt = (struct iscsi_bd *) hba->mp_bd_tbl;
+       addr = (unsigned long) hba->dummy_buf_dma;
+       mp_bdt->buffer_addr_lo = addr & 0xffffffff;
+       mp_bdt->buffer_addr_hi = addr >> 32;
+       mp_bdt->buffer_length = CNIC_PAGE_SIZE;
+       mp_bdt->flags = ISCSI_BD_LAST_IN_BD_CHAIN |
+                       ISCSI_BD_FIRST_IN_BD_CHAIN;
+out:
+       return rc;
+}
+
+
+/**
+ * bnx2i_free_mp_bdt - releases ITT back to free pool
+ * @hba:       pointer to adapter instance
+ *
+ * free MP dummy buffer and associated BD table
+ */
+static void bnx2i_free_mp_bdt(struct bnx2i_hba *hba)
+{
+       if (hba->mp_bd_tbl) {
+               dma_free_coherent(&hba->pcidev->dev, CNIC_PAGE_SIZE,
+                                 hba->mp_bd_tbl, hba->mp_bd_dma);
+               hba->mp_bd_tbl = NULL;
+       }
+       if (hba->dummy_buffer) {
+               dma_free_coherent(&hba->pcidev->dev, CNIC_PAGE_SIZE,
+                                 hba->dummy_buffer, hba->dummy_buf_dma);
+               hba->dummy_buffer = NULL;
+       }
+               return;
+}
+
+/**
+ * bnx2i_drop_session - notifies iscsid of connection error.
+ * @hba:       adapter instance pointer
+ * @session:   iscsi session pointer
+ *
+ * This notifies iscsid that there is a error, so it can initiate
+ * recovery.
+ *
+ * This relies on caller using the iscsi class iterator so the object
+ * is refcounted and does not disapper from under us.
+ */
+void bnx2i_drop_session(struct iscsi_cls_session *cls_session)
+{
+       iscsi_session_failure(cls_session->dd_data, ISCSI_ERR_CONN_FAILED);
+}
+
+/**
+ * bnx2i_ep_destroy_list_add - add an entry to EP destroy list
+ * @hba:       pointer to adapter instance
+ * @ep:                pointer to endpoint (transport identifier) structure
+ *
+ * EP destroy queue manager
+ */
+static int bnx2i_ep_destroy_list_add(struct bnx2i_hba *hba,
+                                    struct bnx2i_endpoint *ep)
+{
+       write_lock_bh(&hba->ep_rdwr_lock);
+       list_add_tail(&ep->link, &hba->ep_destroy_list);
+       write_unlock_bh(&hba->ep_rdwr_lock);
+       return 0;
+}
+
+/**
+ * bnx2i_ep_destroy_list_del - add an entry to EP destroy list
+ *
+ * @hba:               pointer to adapter instance
+ * @ep:                pointer to endpoint (transport identifier) structure
+ *
+ * EP destroy queue manager
+ */
+static int bnx2i_ep_destroy_list_del(struct bnx2i_hba *hba,
+                                    struct bnx2i_endpoint *ep)
+{
+       write_lock_bh(&hba->ep_rdwr_lock);
+       list_del_init(&ep->link);
+       write_unlock_bh(&hba->ep_rdwr_lock);
+
+       return 0;
+}
+
+/**
+ * bnx2i_ep_ofld_list_add - add an entry to ep offload pending list
+ * @hba:       pointer to adapter instance
+ * @ep:                pointer to endpoint (transport identifier) structure
+ *
+ * pending conn offload completion queue manager
+ */
+static int bnx2i_ep_ofld_list_add(struct bnx2i_hba *hba,
+                                 struct bnx2i_endpoint *ep)
+{
+       write_lock_bh(&hba->ep_rdwr_lock);
+       list_add_tail(&ep->link, &hba->ep_ofld_list);
+       write_unlock_bh(&hba->ep_rdwr_lock);
+       return 0;
+}
+
+/**
+ * bnx2i_ep_ofld_list_del - add an entry to ep offload pending list
+ * @hba:               pointer to adapter instance
+ * @ep:                pointer to endpoint (transport identifier) structure
+ *
+ * pending conn offload completion queue manager
+ */
+static int bnx2i_ep_ofld_list_del(struct bnx2i_hba *hba,
+                                 struct bnx2i_endpoint *ep)
+{
+       write_lock_bh(&hba->ep_rdwr_lock);
+       list_del_init(&ep->link);
+       write_unlock_bh(&hba->ep_rdwr_lock);
+       return 0;
+}
+
+
+/**
+ * bnx2i_find_ep_in_ofld_list - find iscsi_cid in pending list of endpoints
+ *
+ * @hba:               pointer to adapter instance
+ * @iscsi_cid:         iscsi context ID to find
+ *
+ */
+struct bnx2i_endpoint *
+bnx2i_find_ep_in_ofld_list(struct bnx2i_hba *hba, u32 iscsi_cid)
+{
+       struct list_head *list;
+       struct list_head *tmp;
+       struct bnx2i_endpoint *ep;
+
+       read_lock_bh(&hba->ep_rdwr_lock);
+       list_for_each_safe(list, tmp, &hba->ep_ofld_list) {
+               ep = (struct bnx2i_endpoint *)list;
+
+               if (ep->ep_iscsi_cid == iscsi_cid)
+                       break;
+               ep = NULL;
+       }
+       read_unlock_bh(&hba->ep_rdwr_lock);
+
+       if (!ep)
+               printk(KERN_ERR "l5 cid %d not found\n", iscsi_cid);
+       return ep;
+}
+
+/**
+ * bnx2i_find_ep_in_destroy_list - find iscsi_cid in destroy list
+ * @hba:               pointer to adapter instance
+ * @iscsi_cid:         iscsi context ID to find
+ *
+ */
+struct bnx2i_endpoint *
+bnx2i_find_ep_in_destroy_list(struct bnx2i_hba *hba, u32 iscsi_cid)
+{
+       struct list_head *list;
+       struct list_head *tmp;
+       struct bnx2i_endpoint *ep;
+
+       read_lock_bh(&hba->ep_rdwr_lock);
+       list_for_each_safe(list, tmp, &hba->ep_destroy_list) {
+               ep = (struct bnx2i_endpoint *)list;
+
+               if (ep->ep_iscsi_cid == iscsi_cid)
+                       break;
+               ep = NULL;
+       }
+       read_unlock_bh(&hba->ep_rdwr_lock);
+
+       if (!ep)
+               printk(KERN_ERR "l5 cid %d not found\n", iscsi_cid);
+
+       return ep;
+}
+
+/**
+ * bnx2i_ep_active_list_add - add an entry to ep active list
+ * @hba:       pointer to adapter instance
+ * @ep:                pointer to endpoint (transport identifier) structure
+ *
+ * current active conn queue manager
+ */
+static void bnx2i_ep_active_list_add(struct bnx2i_hba *hba,
+                                    struct bnx2i_endpoint *ep)
+{
+       write_lock_bh(&hba->ep_rdwr_lock);
+       list_add_tail(&ep->link, &hba->ep_active_list);
+       write_unlock_bh(&hba->ep_rdwr_lock);
+}
+
+
+/**
+ * bnx2i_ep_active_list_del - deletes an entry to ep active list
+ * @hba:       pointer to adapter instance
+ * @ep:                pointer to endpoint (transport identifier) structure
+ *
+ * current active conn queue manager
+ */
+static void bnx2i_ep_active_list_del(struct bnx2i_hba *hba,
+                                    struct bnx2i_endpoint *ep)
+{
+       write_lock_bh(&hba->ep_rdwr_lock);
+       list_del_init(&ep->link);
+       write_unlock_bh(&hba->ep_rdwr_lock);
+}
+
+
+/**
+ * bnx2i_setup_host_queue_size - assigns shost->can_queue param
+ * @hba:       pointer to adapter instance
+ * @shost:     scsi host pointer
+ *
+ * Initializes 'can_queue' parameter based on how many outstanding commands
+ *     the device can handle. Each device 5708/5709/57710 has different
+ *     capabilities
+ */
+static void bnx2i_setup_host_queue_size(struct bnx2i_hba *hba,
+                                       struct Scsi_Host *shost)
+{
+       if (test_bit(BNX2I_NX2_DEV_5708, &hba->cnic_dev_type))
+               shost->can_queue = ISCSI_MAX_CMDS_PER_HBA_5708;
+       else if (test_bit(BNX2I_NX2_DEV_5709, &hba->cnic_dev_type))
+               shost->can_queue = ISCSI_MAX_CMDS_PER_HBA_5709;
+       else if (test_bit(BNX2I_NX2_DEV_57710, &hba->cnic_dev_type))
+               shost->can_queue = ISCSI_MAX_CMDS_PER_HBA_57710;
+       else
+               shost->can_queue = ISCSI_MAX_CMDS_PER_HBA_5708;
+}
+
+
+/**
+ * bnx2i_alloc_hba - allocate and init adapter instance
+ * @cnic:      cnic device pointer
+ *
+ * allocate & initialize adapter structure and call other
+ *     support routines to do per adapter initialization
+ */
+struct bnx2i_hba *bnx2i_alloc_hba(struct cnic_dev *cnic)
+{
+       struct Scsi_Host *shost;
+       struct bnx2i_hba *hba;
+
+       shost = iscsi_host_alloc(&bnx2i_host_template, sizeof(*hba), 0);
+       if (!shost)
+               return NULL;
+       shost->dma_boundary = cnic->pcidev->dma_mask;
+       shost->transportt = bnx2i_scsi_xport_template;
+       shost->max_id = ISCSI_MAX_CONNS_PER_HBA;
+       shost->max_channel = 0;
+       shost->max_lun = 512;
+       shost->max_cmd_len = 16;
+
+       hba = iscsi_host_priv(shost);
+       hba->shost = shost;
+       hba->netdev = cnic->netdev;
+       /* Get PCI related information and update hba struct members */
+       hba->pcidev = cnic->pcidev;
+       pci_dev_get(hba->pcidev);
+       hba->pci_did = hba->pcidev->device;
+       hba->pci_vid = hba->pcidev->vendor;
+       hba->pci_sdid = hba->pcidev->subsystem_device;
+       hba->pci_svid = hba->pcidev->subsystem_vendor;
+       hba->pci_func = PCI_FUNC(hba->pcidev->devfn);
+       hba->pci_devno = PCI_SLOT(hba->pcidev->devfn);
+
+       bnx2i_identify_device(hba, cnic);
+       bnx2i_setup_host_queue_size(hba, shost);
+
+       hba->reg_base = pci_resource_start(hba->pcidev, 0);
+       if (test_bit(BNX2I_NX2_DEV_5709, &hba->cnic_dev_type)) {
+               hba->regview = pci_iomap(hba->pcidev, 0, BNX2_MQ_CONFIG2);
+               if (!hba->regview)
+                       goto ioreg_map_err;
+       } else if (test_bit(BNX2I_NX2_DEV_57710, &hba->cnic_dev_type)) {
+               hba->regview = pci_iomap(hba->pcidev, 0, 4096);
+               if (!hba->regview)
+                       goto ioreg_map_err;
+       }
+
+       if (bnx2i_setup_mp_bdt(hba))
+               goto mp_bdt_mem_err;
+
+       INIT_LIST_HEAD(&hba->ep_ofld_list);
+       INIT_LIST_HEAD(&hba->ep_active_list);
+       INIT_LIST_HEAD(&hba->ep_destroy_list);
+       rwlock_init(&hba->ep_rdwr_lock);
+
+       hba->mtu_supported = BNX2I_MAX_MTU_SUPPORTED;
+
+       /* different values for 5708/5709/57710 */
+       hba->max_active_conns = ISCSI_MAX_CONNS_PER_HBA;
+
+       if (bnx2i_setup_free_cid_que(hba))
+               goto cid_que_err;
+
+       /* SQ/RQ/CQ size can be changed via sysfx interface */
+       if (test_bit(BNX2I_NX2_DEV_57710, &hba->cnic_dev_type)) {
+               if (sq_size && sq_size <= BNX2I_5770X_SQ_WQES_MAX)
+                       hba->max_sqes = sq_size;
+               else
+                       hba->max_sqes = BNX2I_5770X_SQ_WQES_DEFAULT;
+       } else {        /* 5706/5708/5709 */
+               if (sq_size && sq_size <= BNX2I_570X_SQ_WQES_MAX)
+                       hba->max_sqes = sq_size;
+               else
+                       hba->max_sqes = BNX2I_570X_SQ_WQES_DEFAULT;
+       }
+
+       hba->max_rqes = rq_size;
+       hba->max_cqes = hba->max_sqes + rq_size;
+       if (test_bit(BNX2I_NX2_DEV_57710, &hba->cnic_dev_type)) {
+               if (hba->max_cqes > BNX2I_5770X_CQ_WQES_MAX)
+                       hba->max_cqes = BNX2I_5770X_CQ_WQES_MAX;
+       } else if (hba->max_cqes > BNX2I_570X_CQ_WQES_MAX)
+               hba->max_cqes = BNX2I_570X_CQ_WQES_MAX;
+
+       hba->num_ccell = hba->max_sqes / 2;
+
+       spin_lock_init(&hba->lock);
+       mutex_init(&hba->net_dev_lock);
+       init_waitqueue_head(&hba->eh_wait);
+       if (test_bit(BNX2I_NX2_DEV_57710, &hba->cnic_dev_type)) {
+               hba->hba_shutdown_tmo = 30 * HZ;
+               hba->conn_teardown_tmo = 20 * HZ;
+               hba->conn_ctx_destroy_tmo = 6 * HZ;
+       } else {        /* 5706/5708/5709 */
+               hba->hba_shutdown_tmo = 20 * HZ;
+               hba->conn_teardown_tmo = 10 * HZ;
+               hba->conn_ctx_destroy_tmo = 2 * HZ;
+       }
+
+#ifdef CONFIG_32BIT
+       spin_lock_init(&hba->stat_lock);
+#endif
+       memset(&hba->stats, 0, sizeof(struct iscsi_stats_info));
+
+       if (iscsi_host_add(shost, &hba->pcidev->dev))
+               goto free_dump_mem;
+       return hba;
+
+free_dump_mem:
+       bnx2i_release_free_cid_que(hba);
+cid_que_err:
+       bnx2i_free_mp_bdt(hba);
+mp_bdt_mem_err:
+       if (hba->regview) {
+               pci_iounmap(hba->pcidev, hba->regview);
+               hba->regview = NULL;
+       }
+ioreg_map_err:
+       pci_dev_put(hba->pcidev);
+       scsi_host_put(shost);
+       return NULL;
+}
+
+/**
+ * bnx2i_free_hba- releases hba structure and resources held by the adapter
+ * @hba:       pointer to adapter instance
+ *
+ * free adapter structure and call various cleanup routines.
+ */
+void bnx2i_free_hba(struct bnx2i_hba *hba)
+{
+       struct Scsi_Host *shost = hba->shost;
+
+       iscsi_host_remove(shost);
+       INIT_LIST_HEAD(&hba->ep_ofld_list);
+       INIT_LIST_HEAD(&hba->ep_active_list);
+       INIT_LIST_HEAD(&hba->ep_destroy_list);
+       pci_dev_put(hba->pcidev);
+
+       if (hba->regview) {
+               pci_iounmap(hba->pcidev, hba->regview);
+               hba->regview = NULL;
+       }
+       bnx2i_free_mp_bdt(hba);
+       bnx2i_release_free_cid_que(hba);
+       iscsi_host_free(shost);
+}
+
+/**
+ * bnx2i_conn_free_login_resources - free DMA resources used for login process
+ * @hba:               pointer to adapter instance
+ * @bnx2i_conn:                iscsi connection pointer
+ *
+ * Login related resources, mostly BDT & payload DMA memory is freed
+ */
+static void bnx2i_conn_free_login_resources(struct bnx2i_hba *hba,
+                                           struct bnx2i_conn *bnx2i_conn)
+{
+       if (bnx2i_conn->gen_pdu.resp_bd_tbl) {
+               dma_free_coherent(&hba->pcidev->dev, CNIC_PAGE_SIZE,
+                                 bnx2i_conn->gen_pdu.resp_bd_tbl,
+                                 bnx2i_conn->gen_pdu.resp_bd_dma);
+               bnx2i_conn->gen_pdu.resp_bd_tbl = NULL;
+       }
+
+       if (bnx2i_conn->gen_pdu.req_bd_tbl) {
+               dma_free_coherent(&hba->pcidev->dev, CNIC_PAGE_SIZE,
+                                 bnx2i_conn->gen_pdu.req_bd_tbl,
+                                 bnx2i_conn->gen_pdu.req_bd_dma);
+               bnx2i_conn->gen_pdu.req_bd_tbl = NULL;
+       }
+
+       if (bnx2i_conn->gen_pdu.resp_buf) {
+               dma_free_coherent(&hba->pcidev->dev,
+                                 ISCSI_DEF_MAX_RECV_SEG_LEN,
+                                 bnx2i_conn->gen_pdu.resp_buf,
+                                 bnx2i_conn->gen_pdu.resp_dma_addr);
+               bnx2i_conn->gen_pdu.resp_buf = NULL;
+       }
+
+       if (bnx2i_conn->gen_pdu.req_buf) {
+               dma_free_coherent(&hba->pcidev->dev,
+                                 ISCSI_DEF_MAX_RECV_SEG_LEN,
+                                 bnx2i_conn->gen_pdu.req_buf,
+                                 bnx2i_conn->gen_pdu.req_dma_addr);
+               bnx2i_conn->gen_pdu.req_buf = NULL;
+       }
+}
+
+/**
+ * bnx2i_conn_alloc_login_resources - alloc DMA resources for login/nop.
+ * @hba:               pointer to adapter instance
+ * @bnx2i_conn:                iscsi connection pointer
+ *
+ * Mgmt task DNA resources are allocated in this routine.
+ */
+static int bnx2i_conn_alloc_login_resources(struct bnx2i_hba *hba,
+                                           struct bnx2i_conn *bnx2i_conn)
+{
+       /* Allocate memory for login request/response buffers */
+       bnx2i_conn->gen_pdu.req_buf =
+               dma_alloc_coherent(&hba->pcidev->dev,
+                                  ISCSI_DEF_MAX_RECV_SEG_LEN,
+                                  &bnx2i_conn->gen_pdu.req_dma_addr,
+                                  GFP_KERNEL);
+       if (bnx2i_conn->gen_pdu.req_buf == NULL)
+               goto login_req_buf_failure;
+
+       bnx2i_conn->gen_pdu.req_buf_size = 0;
+       bnx2i_conn->gen_pdu.req_wr_ptr = bnx2i_conn->gen_pdu.req_buf;
+
+       bnx2i_conn->gen_pdu.resp_buf =
+               dma_alloc_coherent(&hba->pcidev->dev,
+                                  ISCSI_DEF_MAX_RECV_SEG_LEN,
+                                  &bnx2i_conn->gen_pdu.resp_dma_addr,
+                                  GFP_KERNEL);
+       if (bnx2i_conn->gen_pdu.resp_buf == NULL)
+               goto login_resp_buf_failure;
+
+       bnx2i_conn->gen_pdu.resp_buf_size = ISCSI_DEF_MAX_RECV_SEG_LEN;
+       bnx2i_conn->gen_pdu.resp_wr_ptr = bnx2i_conn->gen_pdu.resp_buf;
+
+       bnx2i_conn->gen_pdu.req_bd_tbl =
+               dma_alloc_coherent(&hba->pcidev->dev, CNIC_PAGE_SIZE,
+                                  &bnx2i_conn->gen_pdu.req_bd_dma, GFP_KERNEL);
+       if (bnx2i_conn->gen_pdu.req_bd_tbl == NULL)
+               goto login_req_bd_tbl_failure;
+
+       bnx2i_conn->gen_pdu.resp_bd_tbl =
+               dma_alloc_coherent(&hba->pcidev->dev, CNIC_PAGE_SIZE,
+                                  &bnx2i_conn->gen_pdu.resp_bd_dma,
+                                  GFP_KERNEL);
+       if (bnx2i_conn->gen_pdu.resp_bd_tbl == NULL)
+               goto login_resp_bd_tbl_failure;
+
+       return 0;
+
+login_resp_bd_tbl_failure:
+       dma_free_coherent(&hba->pcidev->dev, CNIC_PAGE_SIZE,
+                         bnx2i_conn->gen_pdu.req_bd_tbl,
+                         bnx2i_conn->gen_pdu.req_bd_dma);
+       bnx2i_conn->gen_pdu.req_bd_tbl = NULL;
+
+login_req_bd_tbl_failure:
+       dma_free_coherent(&hba->pcidev->dev, ISCSI_DEF_MAX_RECV_SEG_LEN,
+                         bnx2i_conn->gen_pdu.resp_buf,
+                         bnx2i_conn->gen_pdu.resp_dma_addr);
+       bnx2i_conn->gen_pdu.resp_buf = NULL;
+login_resp_buf_failure:
+       dma_free_coherent(&hba->pcidev->dev, ISCSI_DEF_MAX_RECV_SEG_LEN,
+                         bnx2i_conn->gen_pdu.req_buf,
+                         bnx2i_conn->gen_pdu.req_dma_addr);
+       bnx2i_conn->gen_pdu.req_buf = NULL;
+login_req_buf_failure:
+       iscsi_conn_printk(KERN_ERR, bnx2i_conn->cls_conn->dd_data,
+                         "login resource alloc failed!!\n");
+       return -ENOMEM;
+
+}
+
+
+/**
+ * bnx2i_iscsi_prep_generic_pdu_bd - prepares BD table.
+ * @bnx2i_conn:                iscsi connection pointer
+ *
+ * Allocates buffers and BD tables before shipping requests to cnic
+ *     for PDUs prepared by 'iscsid' daemon
+ */
+static void bnx2i_iscsi_prep_generic_pdu_bd(struct bnx2i_conn *bnx2i_conn)
+{
+       struct iscsi_bd *bd_tbl;
+
+       bd_tbl = (struct iscsi_bd *) bnx2i_conn->gen_pdu.req_bd_tbl;
+
+       bd_tbl->buffer_addr_hi =
+               (u32) ((u64) bnx2i_conn->gen_pdu.req_dma_addr >> 32);
+       bd_tbl->buffer_addr_lo = (u32) bnx2i_conn->gen_pdu.req_dma_addr;
+       bd_tbl->buffer_length = bnx2i_conn->gen_pdu.req_wr_ptr -
+                               bnx2i_conn->gen_pdu.req_buf;
+       bd_tbl->reserved0 = 0;
+       bd_tbl->flags = ISCSI_BD_LAST_IN_BD_CHAIN |
+                       ISCSI_BD_FIRST_IN_BD_CHAIN;
+
+       bd_tbl = (struct iscsi_bd  *) bnx2i_conn->gen_pdu.resp_bd_tbl;
+       bd_tbl->buffer_addr_hi = (u64) bnx2i_conn->gen_pdu.resp_dma_addr >> 32;
+       bd_tbl->buffer_addr_lo = (u32) bnx2i_conn->gen_pdu.resp_dma_addr;
+       bd_tbl->buffer_length = ISCSI_DEF_MAX_RECV_SEG_LEN;
+       bd_tbl->reserved0 = 0;
+       bd_tbl->flags = ISCSI_BD_LAST_IN_BD_CHAIN |
+                       ISCSI_BD_FIRST_IN_BD_CHAIN;
+}
+
+
+/**
+ * bnx2i_iscsi_send_generic_request - called to send mgmt tasks.
+ * @task:      transport layer task pointer
+ *
+ * called to transmit PDUs prepared by the 'iscsid' daemon. iSCSI login,
+ *     Nop-out and Logout requests flow through this path.
+ */
+static int bnx2i_iscsi_send_generic_request(struct iscsi_task *task)
+{
+       struct bnx2i_cmd *cmd = task->dd_data;
+       struct bnx2i_conn *bnx2i_conn = cmd->conn;
+       int rc = 0;
+       char *buf;
+       int data_len;
+
+       bnx2i_iscsi_prep_generic_pdu_bd(bnx2i_conn);
+       switch (task->hdr->opcode & ISCSI_OPCODE_MASK) {
+       case ISCSI_OP_LOGIN:
+               bnx2i_send_iscsi_login(bnx2i_conn, task);
+               break;
+       case ISCSI_OP_NOOP_OUT:
+               data_len = bnx2i_conn->gen_pdu.req_buf_size;
+               buf = bnx2i_conn->gen_pdu.req_buf;
+               if (data_len)
+                       rc = bnx2i_send_iscsi_nopout(bnx2i_conn, task,
+                                                    buf, data_len, 1);
+               else
+                       rc = bnx2i_send_iscsi_nopout(bnx2i_conn, task,
+                                                    NULL, 0, 1);
+               break;
+       case ISCSI_OP_LOGOUT:
+               rc = bnx2i_send_iscsi_logout(bnx2i_conn, task);
+               break;
+       case ISCSI_OP_SCSI_TMFUNC:
+               rc = bnx2i_send_iscsi_tmf(bnx2i_conn, task);
+               break;
+       case ISCSI_OP_TEXT:
+               rc = bnx2i_send_iscsi_text(bnx2i_conn, task);
+               break;
+       default:
+               iscsi_conn_printk(KERN_ALERT, bnx2i_conn->cls_conn->dd_data,
+                                 "send_gen: unsupported op 0x%x\n",
+                                 task->hdr->opcode);
+       }
+       return rc;
+}
+
+
+/**********************************************************************
+ *             SCSI-ML Interface
+ **********************************************************************/
+
+/**
+ * bnx2i_cpy_scsi_cdb - copies LUN & CDB fields in required format to sq wqe
+ * @sc:                SCSI-ML command pointer
+ * @cmd:       iscsi cmd pointer
+ */
+static void bnx2i_cpy_scsi_cdb(struct scsi_cmnd *sc, struct bnx2i_cmd *cmd)
+{
+       u32 dword;
+       int lpcnt;
+       u8 *srcp;
+       u32 *dstp;
+       u32 scsi_lun[2];
+
+       int_to_scsilun(sc->device->lun, (struct scsi_lun *) scsi_lun);
+       cmd->req.lun[0] = be32_to_cpu(scsi_lun[0]);
+       cmd->req.lun[1] = be32_to_cpu(scsi_lun[1]);
+
+       lpcnt = cmd->scsi_cmd->cmd_len / sizeof(dword);
+       srcp = (u8 *) sc->cmnd;
+       dstp = (u32 *) cmd->req.cdb;
+       while (lpcnt--) {
+               memcpy(&dword, (const void *) srcp, 4);
+               *dstp = cpu_to_be32(dword);
+               srcp += 4;
+               dstp++;
+       }
+       if (sc->cmd_len & 0x3) {
+               dword = (u32) srcp[0] | ((u32) srcp[1] << 8);
+               *dstp = cpu_to_be32(dword);
+       }
+}
+
+static void bnx2i_cleanup_task(struct iscsi_task *task)
+{
+       struct iscsi_conn *conn = task->conn;
+       struct bnx2i_conn *bnx2i_conn = conn->dd_data;
+       struct bnx2i_hba *hba = bnx2i_conn->hba;
+
+       /*
+        * mgmt task or cmd was never sent to us to transmit.
+        */
+       if (!task->sc || task->state == ISCSI_TASK_PENDING)
+               return;
+       /*
+        * need to clean-up task context to claim dma buffers
+        */
+       if (task->state == ISCSI_TASK_ABRT_TMF) {
+               bnx2i_send_cmd_cleanup_req(hba, task->dd_data);
+
+               spin_unlock_bh(&conn->session->back_lock);
+               wait_for_completion_timeout(&bnx2i_conn->cmd_cleanup_cmpl,
+                               msecs_to_jiffies(ISCSI_CMD_CLEANUP_TIMEOUT));
+               spin_lock_bh(&conn->session->back_lock);
+       }
+       bnx2i_iscsi_unmap_sg_list(task->dd_data);
+}
+
+/**
+ * bnx2i_mtask_xmit - transmit mtask to chip for further processing
+ * @conn:      transport layer conn structure pointer
+ * @task:      transport layer command structure pointer
+ */
+static int
+bnx2i_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task)
+{
+       struct bnx2i_conn *bnx2i_conn = conn->dd_data;
+       struct bnx2i_hba *hba = bnx2i_conn->hba;
+       struct bnx2i_cmd *cmd = task->dd_data;
+
+       memset(bnx2i_conn->gen_pdu.req_buf, 0, ISCSI_DEF_MAX_RECV_SEG_LEN);
+
+       bnx2i_setup_cmd_wqe_template(cmd);
+       bnx2i_conn->gen_pdu.req_buf_size = task->data_count;
+
+       /* Tx PDU/data length count */
+       ADD_STATS_64(hba, tx_pdus, 1);
+       ADD_STATS_64(hba, tx_bytes, task->data_count);
+
+       if (task->data_count) {
+               memcpy(bnx2i_conn->gen_pdu.req_buf, task->data,
+                      task->data_count);
+               bnx2i_conn->gen_pdu.req_wr_ptr =
+                       bnx2i_conn->gen_pdu.req_buf + task->data_count;
+       }
+       cmd->conn = conn->dd_data;
+       cmd->scsi_cmd = NULL;
+       return bnx2i_iscsi_send_generic_request(task);
+}
+
+/**
+ * bnx2i_task_xmit - transmit iscsi command to chip for further processing
+ * @task:      transport layer command structure pointer
+ *
+ * maps SG buffers and send request to chip/firmware in the form of SQ WQE
+ */
+static int bnx2i_task_xmit(struct iscsi_task *task)
+{
+       struct iscsi_conn *conn = task->conn;
+       struct iscsi_session *session = conn->session;
+       struct Scsi_Host *shost = iscsi_session_to_shost(session->cls_session);
+       struct bnx2i_hba *hba = iscsi_host_priv(shost);
+       struct bnx2i_conn *bnx2i_conn = conn->dd_data;
+       struct scsi_cmnd *sc = task->sc;
+       struct bnx2i_cmd *cmd = task->dd_data;
+       struct iscsi_scsi_req *hdr = (struct iscsi_scsi_req *)task->hdr;
+
+       if (atomic_read(&bnx2i_conn->ep->num_active_cmds) + 1  >
+           hba->max_sqes)
+               return -ENOMEM;
+
+       /*
+        * If there is no scsi_cmnd this must be a mgmt task
+        */
+       if (!sc)
+               return bnx2i_mtask_xmit(conn, task);
+
+       bnx2i_setup_cmd_wqe_template(cmd);
+       cmd->req.op_code = ISCSI_OP_SCSI_CMD;
+       cmd->conn = bnx2i_conn;
+       cmd->scsi_cmd = sc;
+       cmd->req.total_data_transfer_length = scsi_bufflen(sc);
+       cmd->req.cmd_sn = be32_to_cpu(hdr->cmdsn);
+
+       bnx2i_iscsi_map_sg_list(cmd);
+       bnx2i_cpy_scsi_cdb(sc, cmd);
+
+       cmd->req.op_attr = ISCSI_ATTR_SIMPLE;
+       if (sc->sc_data_direction == DMA_TO_DEVICE) {
+               cmd->req.op_attr |= ISCSI_CMD_REQUEST_WRITE;
+               cmd->req.itt = task->itt |
+                       (ISCSI_TASK_TYPE_WRITE << ISCSI_CMD_REQUEST_TYPE_SHIFT);
+               bnx2i_setup_write_cmd_bd_info(task);
+       } else {
+               if (scsi_bufflen(sc))
+                       cmd->req.op_attr |= ISCSI_CMD_REQUEST_READ;
+               cmd->req.itt = task->itt |
+                       (ISCSI_TASK_TYPE_READ << ISCSI_CMD_REQUEST_TYPE_SHIFT);
+       }
+
+       cmd->req.num_bds = cmd->io_tbl.bd_valid;
+       if (!cmd->io_tbl.bd_valid) {
+               cmd->req.bd_list_addr_lo = (u32) hba->mp_bd_dma;
+               cmd->req.bd_list_addr_hi = (u32) ((u64) hba->mp_bd_dma >> 32);
+               cmd->req.num_bds = 1;
+       }
+
+       bnx2i_send_iscsi_scsicmd(bnx2i_conn, cmd);
+       return 0;
+}
+
+/**
+ * bnx2i_session_create - create a new iscsi session
+ * @cmds_max:          max commands supported
+ * @qdepth:            scsi queue depth to support
+ * @initial_cmdsn:     initial iscsi CMDSN to be used for this session
+ *
+ * Creates a new iSCSI session instance on given device.
+ */
+static struct iscsi_cls_session *
+bnx2i_session_create(struct iscsi_endpoint *ep,
+                    uint16_t cmds_max, uint16_t qdepth,
+                    uint32_t initial_cmdsn)
+{
+       struct Scsi_Host *shost;
+       struct iscsi_cls_session *cls_session;
+       struct bnx2i_hba *hba;
+       struct bnx2i_endpoint *bnx2i_ep;
+
+       if (!ep) {
+               printk(KERN_ERR "bnx2i: missing ep.\n");
+               return NULL;
+       }
+
+       bnx2i_ep = ep->dd_data;
+       shost = bnx2i_ep->hba->shost;
+       hba = iscsi_host_priv(shost);
+       if (bnx2i_adapter_ready(hba))
+               return NULL;
+
+       /*
+        * user can override hw limit as long as it is within
+        * the min/max.
+        */
+       if (cmds_max > hba->max_sqes)
+               cmds_max = hba->max_sqes;
+       else if (cmds_max < BNX2I_SQ_WQES_MIN)
+               cmds_max = BNX2I_SQ_WQES_MIN;
+
+       cls_session = iscsi_session_setup(&bnx2i_iscsi_transport, shost,
+                                         cmds_max, 0, sizeof(struct bnx2i_cmd),
+                                         initial_cmdsn, ISCSI_MAX_TARGET);
+       if (!cls_session)
+               return NULL;
+
+       if (bnx2i_setup_cmd_pool(hba, cls_session->dd_data))
+               goto session_teardown;
+       return cls_session;
+
+session_teardown:
+       iscsi_session_teardown(cls_session);
+       return NULL;
+}
+
+
+/**
+ * bnx2i_session_destroy - destroys iscsi session
+ * @cls_session:       pointer to iscsi cls session
+ *
+ * Destroys previously created iSCSI session instance and releases
+ *     all resources held by it
+ */
+static void bnx2i_session_destroy(struct iscsi_cls_session *cls_session)
+{
+       struct iscsi_session *session = cls_session->dd_data;
+       struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
+       struct bnx2i_hba *hba = iscsi_host_priv(shost);
+
+       bnx2i_destroy_cmd_pool(hba, session);
+       iscsi_session_teardown(cls_session);
+}
+
+
+/**
+ * bnx2i_conn_create - create iscsi connection instance
+ * @cls_session:       pointer to iscsi cls session
+ * @cid:               iscsi cid as per rfc (not NX2's CID terminology)
+ *
+ * Creates a new iSCSI connection instance for a given session
+ */
+static struct iscsi_cls_conn *
+bnx2i_conn_create(struct iscsi_cls_session *cls_session, uint32_t cid)
+{
+       struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
+       struct bnx2i_hba *hba = iscsi_host_priv(shost);
+       struct bnx2i_conn *bnx2i_conn;
+       struct iscsi_cls_conn *cls_conn;
+       struct iscsi_conn *conn;
+
+       cls_conn = iscsi_conn_setup(cls_session, sizeof(*bnx2i_conn),
+                                   cid);
+       if (!cls_conn)
+               return NULL;
+       conn = cls_conn->dd_data;
+
+       bnx2i_conn = conn->dd_data;
+       bnx2i_conn->cls_conn = cls_conn;
+       bnx2i_conn->hba = hba;
+
+       atomic_set(&bnx2i_conn->work_cnt, 0);
+
+       /* 'ep' ptr will be assigned in bind() call */
+       bnx2i_conn->ep = NULL;
+       init_completion(&bnx2i_conn->cmd_cleanup_cmpl);
+
+       if (bnx2i_conn_alloc_login_resources(hba, bnx2i_conn)) {
+               iscsi_conn_printk(KERN_ALERT, conn,
+                                 "conn_new: login resc alloc failed!!\n");
+               goto free_conn;
+       }
+
+       return cls_conn;
+
+free_conn:
+       iscsi_conn_teardown(cls_conn);
+       return NULL;
+}
+
+/**
+ * bnx2i_conn_bind - binds iscsi sess, conn and ep objects together
+ * @cls_session:       pointer to iscsi cls session
+ * @cls_conn:          pointer to iscsi cls conn
+ * @transport_fd:      64-bit EP handle
+ * @is_leading:                leading connection on this session?
+ *
+ * Binds together iSCSI session instance, iSCSI connection instance
+ *     and the TCP connection. This routine returns error code if
+ *     TCP connection does not belong on the device iSCSI sess/conn
+ *     is bound
+ */
+static int bnx2i_conn_bind(struct iscsi_cls_session *cls_session,
+                          struct iscsi_cls_conn *cls_conn,
+                          uint64_t transport_fd, int is_leading)
+{
+       struct iscsi_conn *conn = cls_conn->dd_data;
+       struct bnx2i_conn *bnx2i_conn = conn->dd_data;
+       struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
+       struct bnx2i_hba *hba = iscsi_host_priv(shost);
+       struct bnx2i_endpoint *bnx2i_ep;
+       struct iscsi_endpoint *ep;
+       int ret_code;
+
+       ep = iscsi_lookup_endpoint(transport_fd);
+       if (!ep)
+               return -EINVAL;
+       /*
+        * Forcefully terminate all in progress connection recovery at the
+        * earliest, either in bind(), send_pdu(LOGIN), or conn_start()
+        */
+       if (bnx2i_adapter_ready(hba))
+               return -EIO;
+
+       bnx2i_ep = ep->dd_data;
+       if ((bnx2i_ep->state == EP_STATE_TCP_FIN_RCVD) ||
+           (bnx2i_ep->state == EP_STATE_TCP_RST_RCVD))
+               /* Peer disconnect via' FIN or RST */
+               return -EINVAL;
+
+       if (iscsi_conn_bind(cls_session, cls_conn, is_leading))
+               return -EINVAL;
+
+       if (bnx2i_ep->hba != hba) {
+               /* Error - TCP connection does not belong to this device
+                */
+               iscsi_conn_printk(KERN_ALERT, cls_conn->dd_data,
+                                 "conn bind, ep=0x%p (%s) does not",
+                                 bnx2i_ep, bnx2i_ep->hba->netdev->name);
+               iscsi_conn_printk(KERN_ALERT, cls_conn->dd_data,
+                                 "belong to hba (%s)\n",
+                                 hba->netdev->name);
+               return -EEXIST;
+       }
+       bnx2i_ep->conn = bnx2i_conn;
+       bnx2i_conn->ep = bnx2i_ep;
+       bnx2i_conn->iscsi_conn_cid = bnx2i_ep->ep_iscsi_cid;
+       bnx2i_conn->fw_cid = bnx2i_ep->ep_cid;
+
+       ret_code = bnx2i_bind_conn_to_iscsi_cid(hba, bnx2i_conn,
+                                               bnx2i_ep->ep_iscsi_cid);
+
+       /* 5706/5708/5709 FW takes RQ as full when initiated, but for 57710
+        * driver needs to explicitly replenish RQ index during setup.
+        */
+       if (test_bit(BNX2I_NX2_DEV_57710, &bnx2i_ep->hba->cnic_dev_type))
+               bnx2i_put_rq_buf(bnx2i_conn, 0);
+
+       bnx2i_arm_cq_event_coalescing(bnx2i_conn->ep, CNIC_ARM_CQE);
+       return ret_code;
+}
+
+
+/**
+ * bnx2i_conn_destroy - destroy iscsi connection instance & release resources
+ * @cls_conn:  pointer to iscsi cls conn
+ *
+ * Destroy an iSCSI connection instance and release memory resources held by
+ *     this connection
+ */
+static void bnx2i_conn_destroy(struct iscsi_cls_conn *cls_conn)
+{
+       struct iscsi_conn *conn = cls_conn->dd_data;
+       struct bnx2i_conn *bnx2i_conn = conn->dd_data;
+       struct Scsi_Host *shost;
+       struct bnx2i_hba *hba;
+       struct bnx2i_work *work, *tmp;
+       unsigned cpu = 0;
+       struct bnx2i_percpu_s *p;
+
+       shost = iscsi_session_to_shost(iscsi_conn_to_session(cls_conn));
+       hba = iscsi_host_priv(shost);
+
+       bnx2i_conn_free_login_resources(hba, bnx2i_conn);
+
+       if (atomic_read(&bnx2i_conn->work_cnt)) {
+               for_each_online_cpu(cpu) {
+                       p = &per_cpu(bnx2i_percpu, cpu);
+                       spin_lock_bh(&p->p_work_lock);
+                       list_for_each_entry_safe(work, tmp,
+                                                &p->work_list, list) {
+                               if (work->session == conn->session &&
+                                   work->bnx2i_conn == bnx2i_conn) {
+                                       list_del_init(&work->list);
+                                       kfree(work);
+                                       if (!atomic_dec_and_test(
+                                                       &bnx2i_conn->work_cnt))
+                                               break;
+                               }
+                       }
+                       spin_unlock_bh(&p->p_work_lock);
+               }
+       }
+
+       iscsi_conn_teardown(cls_conn);
+}
+
+
+/**
+ * bnx2i_ep_get_param - return iscsi ep parameter to caller
+ * @ep:                pointer to iscsi endpoint
+ * @param:     parameter type identifier
+ * @buf:       buffer pointer
+ *
+ * returns iSCSI ep parameters
+ */
+static int bnx2i_ep_get_param(struct iscsi_endpoint *ep,
+                             enum iscsi_param param, char *buf)
+{
+       struct bnx2i_endpoint *bnx2i_ep = ep->dd_data;
+       struct bnx2i_hba *hba = bnx2i_ep->hba;
+       int len = -ENOTCONN;
+
+       if (!hba)
+               return -ENOTCONN;
+
+       switch (param) {
+       case ISCSI_PARAM_CONN_PORT:
+               mutex_lock(&hba->net_dev_lock);
+               if (bnx2i_ep->cm_sk)
+                       len = sprintf(buf, "%hu\n", bnx2i_ep->cm_sk->dst_port);
+               mutex_unlock(&hba->net_dev_lock);
+               break;
+       case ISCSI_PARAM_CONN_ADDRESS:
+               mutex_lock(&hba->net_dev_lock);
+               if (bnx2i_ep->cm_sk)
+                       len = sprintf(buf, "%pI4\n", &bnx2i_ep->cm_sk->dst_ip);
+               mutex_unlock(&hba->net_dev_lock);
+               break;
+       default:
+               return -ENOSYS;
+       }
+
+       return len;
+}
+
+/**
+ * bnx2i_host_get_param - returns host (adapter) related parameters
+ * @shost:     scsi host pointer
+ * @param:     parameter type identifier
+ * @buf:       buffer pointer
+ */
+static int bnx2i_host_get_param(struct Scsi_Host *shost,
+                               enum iscsi_host_param param, char *buf)
+{
+       struct bnx2i_hba *hba = iscsi_host_priv(shost);
+       int len = 0;
+
+       switch (param) {
+       case ISCSI_HOST_PARAM_HWADDRESS:
+               len = sysfs_format_mac(buf, hba->cnic->mac_addr, 6);
+               break;
+       case ISCSI_HOST_PARAM_NETDEV_NAME:
+               len = sprintf(buf, "%s\n", hba->netdev->name);
+               break;
+       case ISCSI_HOST_PARAM_IPADDRESS: {
+               struct list_head *active_list = &hba->ep_active_list;
+
+               read_lock_bh(&hba->ep_rdwr_lock);
+               if (!list_empty(&hba->ep_active_list)) {
+                       struct bnx2i_endpoint *bnx2i_ep;
+                       struct cnic_sock *csk;
+
+                       bnx2i_ep = list_first_entry(active_list,
+                                                   struct bnx2i_endpoint,
+                                                   link);
+                       csk = bnx2i_ep->cm_sk;
+                       if (test_bit(SK_F_IPV6, &csk->flags))
+                               len = sprintf(buf, "%pI6\n", csk->src_ip);
+                       else
+                               len = sprintf(buf, "%pI4\n", csk->src_ip);
+               }
+               read_unlock_bh(&hba->ep_rdwr_lock);
+               break;
+       }
+       default:
+               return iscsi_host_get_param(shost, param, buf);
+       }
+       return len;
+}
+
+/**
+ * bnx2i_conn_start - completes iscsi connection migration to FFP
+ * @cls_conn:  pointer to iscsi cls conn
+ *
+ * last call in FFP migration to handover iscsi conn to the driver
+ */
+static int bnx2i_conn_start(struct iscsi_cls_conn *cls_conn)
+{
+       struct iscsi_conn *conn = cls_conn->dd_data;
+       struct bnx2i_conn *bnx2i_conn = conn->dd_data;
+
+       bnx2i_conn->ep->state = EP_STATE_ULP_UPDATE_START;
+       bnx2i_update_iscsi_conn(conn);
+
+       /*
+        * this should normally not sleep for a long time so it should
+        * not disrupt the caller.
+        */
+       bnx2i_conn->ep->ofld_timer.expires = 1 * HZ + jiffies;
+       bnx2i_conn->ep->ofld_timer.function = bnx2i_ep_ofld_timer;
+       bnx2i_conn->ep->ofld_timer.data = (unsigned long) bnx2i_conn->ep;
+       add_timer(&bnx2i_conn->ep->ofld_timer);
+       /* update iSCSI context for this conn, wait for CNIC to complete */
+       wait_event_interruptible(bnx2i_conn->ep->ofld_wait,
+                       bnx2i_conn->ep->state != EP_STATE_ULP_UPDATE_START);
+
+       if (signal_pending(current))
+               flush_signals(current);
+       del_timer_sync(&bnx2i_conn->ep->ofld_timer);
+
+       iscsi_conn_start(cls_conn);
+       return 0;
+}
+
+
+/**
+ * bnx2i_conn_get_stats - returns iSCSI stats
+ * @cls_conn:  pointer to iscsi cls conn
+ * @stats:     pointer to iscsi statistic struct
+ */
+static void bnx2i_conn_get_stats(struct iscsi_cls_conn *cls_conn,
+                                struct iscsi_stats *stats)
+{
+       struct iscsi_conn *conn = cls_conn->dd_data;
+
+       stats->txdata_octets = conn->txdata_octets;
+       stats->rxdata_octets = conn->rxdata_octets;
+       stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
+       stats->dataout_pdus = conn->dataout_pdus_cnt;
+       stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
+       stats->datain_pdus = conn->datain_pdus_cnt;
+       stats->r2t_pdus = conn->r2t_pdus_cnt;
+       stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
+       stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
+       stats->digest_err = 0;
+       stats->timeout_err = 0;
+       strcpy(stats->custom[0].desc, "eh_abort_cnt");
+       stats->custom[0].value = conn->eh_abort_cnt;
+       stats->custom_length = 1;
+}
+
+
+/**
+ * bnx2i_check_route - checks if target IP route belongs to one of NX2 devices
+ * @dst_addr:  target IP address
+ *
+ * check if route resolves to BNX2 device
+ */
+static struct bnx2i_hba *bnx2i_check_route(struct sockaddr *dst_addr)
+{
+       struct sockaddr_in *desti = (struct sockaddr_in *) dst_addr;
+       struct bnx2i_hba *hba;
+       struct cnic_dev *cnic = NULL;
+
+       hba = get_adapter_list_head();
+       if (hba && hba->cnic)
+               cnic = hba->cnic->cm_select_dev(desti, CNIC_ULP_ISCSI);
+       if (!cnic) {
+               printk(KERN_ALERT "bnx2i: no route,"
+                      "can't connect using cnic\n");
+               goto no_nx2_route;
+       }
+       hba = bnx2i_find_hba_for_cnic(cnic);
+       if (!hba)
+               goto no_nx2_route;
+
+       if (bnx2i_adapter_ready(hba)) {
+               printk(KERN_ALERT "bnx2i: check route, hba not found\n");
+               goto no_nx2_route;
+       }
+       if (hba->netdev->mtu > hba->mtu_supported) {
+               printk(KERN_ALERT "bnx2i: %s network i/f mtu is set to %d\n",
+                                 hba->netdev->name, hba->netdev->mtu);
+               printk(KERN_ALERT "bnx2i: iSCSI HBA can support mtu of %d\n",
+                                 hba->mtu_supported);
+               goto no_nx2_route;
+       }
+       return hba;
+no_nx2_route:
+       return NULL;
+}
+
+
+/**
+ * bnx2i_tear_down_conn - tear down iscsi/tcp connection and free resources
+ * @hba:       pointer to adapter instance
+ * @ep:                endpoint (transport identifier) structure
+ *
+ * destroys cm_sock structure and on chip iscsi context
+ */
+static int bnx2i_tear_down_conn(struct bnx2i_hba *hba,
+                                struct bnx2i_endpoint *ep)
+{
+       if (test_bit(BNX2I_CNIC_REGISTERED, &hba->reg_with_cnic) && ep->cm_sk)
+               hba->cnic->cm_destroy(ep->cm_sk);
+
+       if (test_bit(BNX2I_NX2_DEV_57710, &hba->cnic_dev_type) &&
+           ep->state == EP_STATE_DISCONN_TIMEDOUT) {
+               if (ep->conn && ep->conn->cls_conn &&
+                   ep->conn->cls_conn->dd_data) {
+                       struct iscsi_conn *conn = ep->conn->cls_conn->dd_data;
+
+                       /* Must suspend all rx queue activity for this ep */
+                       set_bit(ISCSI_SUSPEND_BIT, &conn->suspend_rx);
+               }
+               /* CONN_DISCONNECT timeout may or may not be an issue depending
+                * on what transcribed in TCP layer, different targets behave
+                * differently
+                */
+               printk(KERN_ALERT "bnx2i (%s): - WARN - CONN_DISCON timed out, "
+                                 "please submit GRC Dump, NW/PCIe trace, "
+                                 "driver msgs to developers for analysis\n",
+                                 hba->netdev->name);
+       }
+
+       ep->state = EP_STATE_CLEANUP_START;
+       init_timer(&ep->ofld_timer);
+       ep->ofld_timer.expires = hba->conn_ctx_destroy_tmo + jiffies;
+       ep->ofld_timer.function = bnx2i_ep_ofld_timer;
+       ep->ofld_timer.data = (unsigned long) ep;
+       add_timer(&ep->ofld_timer);
+
+       bnx2i_ep_destroy_list_add(hba, ep);
+
+       /* destroy iSCSI context, wait for it to complete */
+       if (bnx2i_send_conn_destroy(hba, ep))
+               ep->state = EP_STATE_CLEANUP_CMPL;
+
+       wait_event_interruptible(ep->ofld_wait,
+                                (ep->state != EP_STATE_CLEANUP_START));
+
+       if (signal_pending(current))
+               flush_signals(current);
+       del_timer_sync(&ep->ofld_timer);
+
+       bnx2i_ep_destroy_list_del(hba, ep);
+
+       if (ep->state != EP_STATE_CLEANUP_CMPL)
+               /* should never happen */
+               printk(KERN_ALERT "bnx2i - conn destroy failed\n");
+
+       return 0;
+}
+
+
+/**
+ * bnx2i_ep_connect - establish TCP connection to target portal
+ * @shost:             scsi host
+ * @dst_addr:          target IP address
+ * @non_blocking:      blocking or non-blocking call
+ *
+ * this routine initiates the TCP/IP connection by invoking Option-2 i/f
+ *     with l5_core and the CNIC. This is a multi-step process of resolving
+ *     route to target, create a iscsi connection context, handshaking with
+ *     CNIC module to create/initialize the socket struct and finally
+ *     sending down option-2 request to complete TCP 3-way handshake
+ */
+static struct iscsi_endpoint *bnx2i_ep_connect(struct Scsi_Host *shost,
+                                              struct sockaddr *dst_addr,
+                                              int non_blocking)
+{
+       u32 iscsi_cid = BNX2I_CID_RESERVED;
+       struct sockaddr_in *desti = (struct sockaddr_in *) dst_addr;
+       struct sockaddr_in6 *desti6;
+       struct bnx2i_endpoint *bnx2i_ep;
+       struct bnx2i_hba *hba;
+       struct cnic_dev *cnic;
+       struct cnic_sockaddr saddr;
+       struct iscsi_endpoint *ep;
+       int rc = 0;
+
+       if (shost) {
+               /* driver is given scsi host to work with */
+               hba = iscsi_host_priv(shost);
+       } else
+               /*
+                * check if the given destination can be reached through
+                * a iscsi capable NetXtreme2 device
+                */
+               hba = bnx2i_check_route(dst_addr);
+
+       if (!hba) {
+               rc = -EINVAL;
+               goto nohba;
+       }
+       mutex_lock(&hba->net_dev_lock);
+
+       if (bnx2i_adapter_ready(hba) || !hba->cid_que.cid_free_cnt) {
+               rc = -EPERM;
+               goto check_busy;
+       }
+       cnic = hba->cnic;
+       ep = bnx2i_alloc_ep(hba);
+       if (!ep) {
+               rc = -ENOMEM;
+               goto check_busy;
+       }
+       bnx2i_ep = ep->dd_data;
+
+       atomic_set(&bnx2i_ep->num_active_cmds, 0);
+       iscsi_cid = bnx2i_alloc_iscsi_cid(hba);
+       if (iscsi_cid == -1) {
+               printk(KERN_ALERT "bnx2i (%s): alloc_ep - unable to allocate "
+                       "iscsi cid\n", hba->netdev->name);
+               rc = -ENOMEM;
+               bnx2i_free_ep(ep);
+               goto check_busy;
+       }
+       bnx2i_ep->hba_age = hba->age;
+
+       rc = bnx2i_alloc_qp_resc(hba, bnx2i_ep);
+       if (rc != 0) {
+               printk(KERN_ALERT "bnx2i (%s): ep_conn - alloc QP resc error"
+                       "\n", hba->netdev->name);
+               rc = -ENOMEM;
+               goto qp_resc_err;
+       }
+
+       bnx2i_ep->ep_iscsi_cid = (u16)iscsi_cid;
+       bnx2i_ep->state = EP_STATE_OFLD_START;
+       bnx2i_ep_ofld_list_add(hba, bnx2i_ep);
+
+       init_timer(&bnx2i_ep->ofld_timer);
+       bnx2i_ep->ofld_timer.expires = 2 * HZ + jiffies;
+       bnx2i_ep->ofld_timer.function = bnx2i_ep_ofld_timer;
+       bnx2i_ep->ofld_timer.data = (unsigned long) bnx2i_ep;
+       add_timer(&bnx2i_ep->ofld_timer);
+
+       if (bnx2i_send_conn_ofld_req(hba, bnx2i_ep)) {
+               if (bnx2i_ep->state == EP_STATE_OFLD_FAILED_CID_BUSY) {
+                       printk(KERN_ALERT "bnx2i (%s): iscsi cid %d is busy\n",
+                               hba->netdev->name, bnx2i_ep->ep_iscsi_cid);
+                       rc = -EBUSY;
+               } else
+                       rc = -ENOSPC;
+               printk(KERN_ALERT "bnx2i (%s): unable to send conn offld kwqe"
+                       "\n", hba->netdev->name);
+               bnx2i_ep_ofld_list_del(hba, bnx2i_ep);
+               goto conn_failed;
+       }
+
+       /* Wait for CNIC hardware to setup conn context and return 'cid' */
+       wait_event_interruptible(bnx2i_ep->ofld_wait,
+                                bnx2i_ep->state != EP_STATE_OFLD_START);
+
+       if (signal_pending(current))
+               flush_signals(current);
+       del_timer_sync(&bnx2i_ep->ofld_timer);
+
+       bnx2i_ep_ofld_list_del(hba, bnx2i_ep);
+
+       if (bnx2i_ep->state != EP_STATE_OFLD_COMPL) {
+               if (bnx2i_ep->state == EP_STATE_OFLD_FAILED_CID_BUSY) {
+                       printk(KERN_ALERT "bnx2i (%s): iscsi cid %d is busy\n",
+                               hba->netdev->name, bnx2i_ep->ep_iscsi_cid);
+                       rc = -EBUSY;
+               } else
+                       rc = -ENOSPC;
+               goto conn_failed;
+       }
+
+       rc = cnic->cm_create(cnic, CNIC_ULP_ISCSI, bnx2i_ep->ep_cid,
+                            iscsi_cid, &bnx2i_ep->cm_sk, bnx2i_ep);
+       if (rc) {
+               rc = -EINVAL;
+               /* Need to terminate and cleanup the connection */
+               goto release_ep;
+       }
+
+       bnx2i_ep->cm_sk->rcv_buf = 256 * 1024;
+       bnx2i_ep->cm_sk->snd_buf = 256 * 1024;
+       clear_bit(SK_TCP_TIMESTAMP, &bnx2i_ep->cm_sk->tcp_flags);
+
+       memset(&saddr, 0, sizeof(saddr));
+       if (dst_addr->sa_family == AF_INET) {
+               desti = (struct sockaddr_in *) dst_addr;
+               saddr.remote.v4 = *desti;
+               saddr.local.v4.sin_family = desti->sin_family;
+       } else if (dst_addr->sa_family == AF_INET6) {
+               desti6 = (struct sockaddr_in6 *) dst_addr;
+               saddr.remote.v6 = *desti6;
+               saddr.local.v6.sin6_family = desti6->sin6_family;
+       }
+
+       bnx2i_ep->timestamp = jiffies;
+       bnx2i_ep->state = EP_STATE_CONNECT_START;
+       if (!test_bit(BNX2I_CNIC_REGISTERED, &hba->reg_with_cnic)) {
+               rc = -EINVAL;
+               goto conn_failed;
+       } else
+               rc = cnic->cm_connect(bnx2i_ep->cm_sk, &saddr);
+       if (rc)
+               goto release_ep;
+
+       bnx2i_ep_active_list_add(hba, bnx2i_ep);
+
+       if (bnx2i_map_ep_dbell_regs(bnx2i_ep))
+               goto del_active_ep;
+
+       mutex_unlock(&hba->net_dev_lock);
+       return ep;
+
+del_active_ep:
+       bnx2i_ep_active_list_del(hba, bnx2i_ep);
+release_ep:
+       if (bnx2i_tear_down_conn(hba, bnx2i_ep)) {
+               mutex_unlock(&hba->net_dev_lock);
+               return ERR_PTR(rc);
+       }
+conn_failed:
+       bnx2i_free_qp_resc(hba, bnx2i_ep);
+qp_resc_err:
+       bnx2i_free_ep(ep);
+check_busy:
+       mutex_unlock(&hba->net_dev_lock);
+nohba:
+       return ERR_PTR(rc);
+}
+
+
+/**
+ * bnx2i_ep_poll - polls for TCP connection establishement
+ * @ep:                        TCP connection (endpoint) handle
+ * @timeout_ms:                timeout value in milli secs
+ *
+ * polls for TCP connect request to complete
+ */
+static int bnx2i_ep_poll(struct iscsi_endpoint *ep, int timeout_ms)
+{
+       struct bnx2i_endpoint *bnx2i_ep;
+       int rc = 0;
+
+       bnx2i_ep = ep->dd_data;
+       if ((bnx2i_ep->state == EP_STATE_IDLE) ||
+           (bnx2i_ep->state == EP_STATE_CONNECT_FAILED) ||
+           (bnx2i_ep->state == EP_STATE_OFLD_FAILED))
+               return -1;
+       if (bnx2i_ep->state == EP_STATE_CONNECT_COMPL)
+               return 1;
+
+       rc = wait_event_interruptible_timeout(bnx2i_ep->ofld_wait,
+                                             ((bnx2i_ep->state ==
+                                               EP_STATE_OFLD_FAILED) ||
+                                              (bnx2i_ep->state ==
+                                               EP_STATE_CONNECT_FAILED) ||
+                                              (bnx2i_ep->state ==
+                                               EP_STATE_CONNECT_COMPL)),
+                                             msecs_to_jiffies(timeout_ms));
+       if (bnx2i_ep->state == EP_STATE_OFLD_FAILED)
+               rc = -1;
+
+       if (rc > 0)
+               return 1;
+       else if (!rc)
+               return 0;       /* timeout */
+       else
+               return rc;
+}
+
+
+/**
+ * bnx2i_ep_tcp_conn_active - check EP state transition
+ * @ep:                endpoint pointer
+ *
+ * check if underlying TCP connection is active
+ */
+static int bnx2i_ep_tcp_conn_active(struct bnx2i_endpoint *bnx2i_ep)
+{
+       int ret;
+       int cnic_dev_10g = 0;
+
+       if (test_bit(BNX2I_NX2_DEV_57710, &bnx2i_ep->hba->cnic_dev_type))
+               cnic_dev_10g = 1;
+
+       switch (bnx2i_ep->state) {
+       case EP_STATE_CLEANUP_FAILED:
+       case EP_STATE_OFLD_FAILED:
+       case EP_STATE_DISCONN_TIMEDOUT:
+               ret = 0;
+               break;
+       case EP_STATE_CONNECT_START:
+       case EP_STATE_CONNECT_FAILED:
+       case EP_STATE_CONNECT_COMPL:
+       case EP_STATE_ULP_UPDATE_START:
+       case EP_STATE_ULP_UPDATE_COMPL:
+       case EP_STATE_TCP_FIN_RCVD:
+       case EP_STATE_LOGOUT_SENT:
+       case EP_STATE_LOGOUT_RESP_RCVD:
+       case EP_STATE_ULP_UPDATE_FAILED:
+               ret = 1;
+               break;
+       case EP_STATE_TCP_RST_RCVD:
+               if (cnic_dev_10g)
+                       ret = 0;
+               else
+                       ret = 1;
+               break;
+       default:
+               ret = 0;
+       }
+
+       return ret;
+}
+
+
+/*
+ * bnx2i_hw_ep_disconnect - executes TCP connection teardown process in the hw
+ * @ep:                TCP connection (bnx2i endpoint) handle
+ *
+ * executes  TCP connection teardown process
+ */
+int bnx2i_hw_ep_disconnect(struct bnx2i_endpoint *bnx2i_ep)
+{
+       struct bnx2i_hba *hba = bnx2i_ep->hba;
+       struct cnic_dev *cnic;
+       struct iscsi_session *session = NULL;
+       struct iscsi_conn *conn = NULL;
+       int ret = 0;
+       int close = 0;
+       int close_ret = 0;
+
+       if (!hba)
+               return 0;
+
+       cnic = hba->cnic;
+       if (!cnic)
+               return 0;
+
+       if (bnx2i_ep->state == EP_STATE_IDLE ||
+           bnx2i_ep->state == EP_STATE_DISCONN_TIMEDOUT)
+               return 0;
+
+       if (!bnx2i_ep_tcp_conn_active(bnx2i_ep))
+               goto destroy_conn;
+
+       if (bnx2i_ep->conn) {
+               conn = bnx2i_ep->conn->cls_conn->dd_data;
+               session = conn->session;
+       }
+
+       init_timer(&bnx2i_ep->ofld_timer);
+       bnx2i_ep->ofld_timer.expires = hba->conn_teardown_tmo + jiffies;
+       bnx2i_ep->ofld_timer.function = bnx2i_ep_ofld_timer;
+       bnx2i_ep->ofld_timer.data = (unsigned long) bnx2i_ep;
+       add_timer(&bnx2i_ep->ofld_timer);
+
+       if (!test_bit(BNX2I_CNIC_REGISTERED, &hba->reg_with_cnic))
+               goto out;
+
+       if (session) {
+               spin_lock_bh(&session->frwd_lock);
+               if (bnx2i_ep->state != EP_STATE_TCP_FIN_RCVD) {
+                       if (session->state == ISCSI_STATE_LOGGING_OUT) {
+                               if (bnx2i_ep->state == EP_STATE_LOGOUT_SENT) {
+                                       /* Logout sent, but no resp */
+                                       printk(KERN_ALERT "bnx2i (%s): WARNING"
+                                               " logout response was not "
+                                               "received!\n",
+                                               bnx2i_ep->hba->netdev->name);
+                               } else if (bnx2i_ep->state ==
+                                          EP_STATE_LOGOUT_RESP_RCVD)
+                                       close = 1;
+                       }
+               } else
+                       close = 1;
+
+               spin_unlock_bh(&session->frwd_lock);
+       }
+
+       bnx2i_ep->state = EP_STATE_DISCONN_START;
+
+       if (close)
+               close_ret = cnic->cm_close(bnx2i_ep->cm_sk);
+       else
+               close_ret = cnic->cm_abort(bnx2i_ep->cm_sk);
+
+       if (close_ret)
+               printk(KERN_ALERT "bnx2i (%s): close/abort(%d) returned %d\n",
+                       bnx2i_ep->hba->netdev->name, close, close_ret);
+       else
+               /* wait for option-2 conn teardown */
+               wait_event_interruptible(bnx2i_ep->ofld_wait,
+                                bnx2i_ep->state != EP_STATE_DISCONN_START);
+
+       if (signal_pending(current))
+               flush_signals(current);
+       del_timer_sync(&bnx2i_ep->ofld_timer);
+
+destroy_conn:
+       bnx2i_ep_active_list_del(hba, bnx2i_ep);
+       if (bnx2i_tear_down_conn(hba, bnx2i_ep))
+               return -EINVAL;
+out:
+       bnx2i_ep->state = EP_STATE_IDLE;
+       return ret;
+}
+
+
+/**
+ * bnx2i_ep_disconnect - executes TCP connection teardown process
+ * @ep:                TCP connection (iscsi endpoint) handle
+ *
+ * executes  TCP connection teardown process
+ */
+static void bnx2i_ep_disconnect(struct iscsi_endpoint *ep)
+{
+       struct bnx2i_endpoint *bnx2i_ep;
+       struct bnx2i_conn *bnx2i_conn = NULL;
+       struct iscsi_conn *conn = NULL;
+       struct bnx2i_hba *hba;
+
+       bnx2i_ep = ep->dd_data;
+
+       /* driver should not attempt connection cleanup until TCP_CONNECT
+        * completes either successfully or fails. Timeout is 9-secs, so
+        * wait for it to complete
+        */
+       while ((bnx2i_ep->state == EP_STATE_CONNECT_START) &&
+               !time_after(jiffies, bnx2i_ep->timestamp + (12 * HZ)))
+               msleep(250);
+
+       if (bnx2i_ep->conn) {
+               bnx2i_conn = bnx2i_ep->conn;
+               conn = bnx2i_conn->cls_conn->dd_data;
+               iscsi_suspend_queue(conn);
+       }
+       hba = bnx2i_ep->hba;
+
+       mutex_lock(&hba->net_dev_lock);
+
+       if (bnx2i_ep->state == EP_STATE_DISCONN_TIMEDOUT)
+               goto out;
+
+       if (bnx2i_ep->state == EP_STATE_IDLE)
+               goto free_resc;
+
+       if (!test_bit(ADAPTER_STATE_UP, &hba->adapter_state) ||
+           (bnx2i_ep->hba_age != hba->age)) {
+               bnx2i_ep_active_list_del(hba, bnx2i_ep);
+               goto free_resc;
+       }
+
+       /* Do all chip cleanup here */
+       if (bnx2i_hw_ep_disconnect(bnx2i_ep)) {
+               mutex_unlock(&hba->net_dev_lock);
+               return;
+       }
+free_resc:
+       bnx2i_free_qp_resc(hba, bnx2i_ep);
+
+       if (bnx2i_conn)
+               bnx2i_conn->ep = NULL;
+
+       bnx2i_free_ep(ep);
+out:
+       mutex_unlock(&hba->net_dev_lock);
+
+       wake_up_interruptible(&hba->eh_wait);
+}
+
+
+/**
+ * bnx2i_nl_set_path - ISCSI_UEVENT_PATH_UPDATE user message handler
+ * @buf:       pointer to buffer containing iscsi path message
+ *
+ */
+static int bnx2i_nl_set_path(struct Scsi_Host *shost, struct iscsi_path *params)
+{
+       struct bnx2i_hba *hba = iscsi_host_priv(shost);
+       char *buf = (char *) params;
+       u16 len = sizeof(*params);
+
+       /* handled by cnic driver */
+       hba->cnic->iscsi_nl_msg_recv(hba->cnic, ISCSI_UEVENT_PATH_UPDATE, buf,
+                                    len);
+
+       return 0;
+}
+
+static umode_t bnx2i_attr_is_visible(int param_type, int param)
+{
+       switch (param_type) {
+       case ISCSI_HOST_PARAM:
+               switch (param) {
+               case ISCSI_HOST_PARAM_NETDEV_NAME:
+               case ISCSI_HOST_PARAM_HWADDRESS:
+               case ISCSI_HOST_PARAM_IPADDRESS:
+                       return S_IRUGO;
+               default:
+                       return 0;
+               }
+       case ISCSI_PARAM:
+               switch (param) {
+               case ISCSI_PARAM_MAX_RECV_DLENGTH:
+               case ISCSI_PARAM_MAX_XMIT_DLENGTH:
+               case ISCSI_PARAM_HDRDGST_EN:
+               case ISCSI_PARAM_DATADGST_EN:
+               case ISCSI_PARAM_CONN_ADDRESS:
+               case ISCSI_PARAM_CONN_PORT:
+               case ISCSI_PARAM_EXP_STATSN:
+               case ISCSI_PARAM_PERSISTENT_ADDRESS:
+               case ISCSI_PARAM_PERSISTENT_PORT:
+               case ISCSI_PARAM_PING_TMO:
+               case ISCSI_PARAM_RECV_TMO:
+               case ISCSI_PARAM_INITIAL_R2T_EN:
+               case ISCSI_PARAM_MAX_R2T:
+               case ISCSI_PARAM_IMM_DATA_EN:
+               case ISCSI_PARAM_FIRST_BURST:
+               case ISCSI_PARAM_MAX_BURST:
+               case ISCSI_PARAM_PDU_INORDER_EN:
+               case ISCSI_PARAM_DATASEQ_INORDER_EN:
+               case ISCSI_PARAM_ERL:
+               case ISCSI_PARAM_TARGET_NAME:
+               case ISCSI_PARAM_TPGT:
+               case ISCSI_PARAM_USERNAME:
+               case ISCSI_PARAM_PASSWORD:
+               case ISCSI_PARAM_USERNAME_IN:
+               case ISCSI_PARAM_PASSWORD_IN:
+               case ISCSI_PARAM_FAST_ABORT:
+               case ISCSI_PARAM_ABORT_TMO:
+               case ISCSI_PARAM_LU_RESET_TMO:
+               case ISCSI_PARAM_TGT_RESET_TMO:
+               case ISCSI_PARAM_IFACE_NAME:
+               case ISCSI_PARAM_INITIATOR_NAME:
+               case ISCSI_PARAM_BOOT_ROOT:
+               case ISCSI_PARAM_BOOT_NIC:
+               case ISCSI_PARAM_BOOT_TARGET:
+                       return S_IRUGO;
+               default:
+                       return 0;
+               }
+       }
+
+       return 0;
+}
+
+/*
+ * 'Scsi_Host_Template' structure and 'iscsi_tranport' structure template
+ * used while registering with the scsi host and iSCSI transport module.
+ */
+static struct scsi_host_template bnx2i_host_template = {
+       .module                 = THIS_MODULE,
+       .name                   = "QLogic Offload iSCSI Initiator",
+       .proc_name              = "bnx2i",
+       .queuecommand           = iscsi_queuecommand,
+       .eh_abort_handler       = iscsi_eh_abort,
+       .eh_device_reset_handler = iscsi_eh_device_reset,
+       .eh_target_reset_handler = iscsi_eh_recover_target,
+       .change_queue_depth     = scsi_change_queue_depth,
+       .target_alloc           = iscsi_target_alloc,
+       .can_queue              = 2048,
+       .max_sectors            = 127,
+       .cmd_per_lun            = 128,
+       .this_id                = -1,
+       .use_clustering         = ENABLE_CLUSTERING,
+       .sg_tablesize           = ISCSI_MAX_BDS_PER_CMD,
+       .shost_attrs            = bnx2i_dev_attributes,
+       .track_queue_depth      = 1,
+};
+
+struct iscsi_transport bnx2i_iscsi_transport = {
+       .owner                  = THIS_MODULE,
+       .name                   = "bnx2i",
+       .caps                   = CAP_RECOVERY_L0 | CAP_HDRDGST |
+                                 CAP_MULTI_R2T | CAP_DATADGST |
+                                 CAP_DATA_PATH_OFFLOAD |
+                                 CAP_TEXT_NEGO,
+       .create_session         = bnx2i_session_create,
+       .destroy_session        = bnx2i_session_destroy,
+       .create_conn            = bnx2i_conn_create,
+       .bind_conn              = bnx2i_conn_bind,
+       .destroy_conn           = bnx2i_conn_destroy,
+       .attr_is_visible        = bnx2i_attr_is_visible,
+       .set_param              = iscsi_set_param,
+       .get_conn_param         = iscsi_conn_get_param,
+       .get_session_param      = iscsi_session_get_param,
+       .get_host_param         = bnx2i_host_get_param,
+       .start_conn             = bnx2i_conn_start,
+       .stop_conn              = iscsi_conn_stop,
+       .send_pdu               = iscsi_conn_send_pdu,
+       .xmit_task              = bnx2i_task_xmit,
+       .get_stats              = bnx2i_conn_get_stats,
+       /* TCP connect - disconnect - option-2 interface calls */
+       .get_ep_param           = bnx2i_ep_get_param,
+       .ep_connect             = bnx2i_ep_connect,
+       .ep_poll                = bnx2i_ep_poll,
+       .ep_disconnect          = bnx2i_ep_disconnect,
+       .set_path               = bnx2i_nl_set_path,
+       /* Error recovery timeout call */
+       .session_recovery_timedout = iscsi_session_recovery_timedout,
+       .cleanup_task           = bnx2i_cleanup_task,
+};