Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / scsi / cxgbi / cxgb3i / cxgb3i.c
diff --git a/kernel/drivers/scsi/cxgbi/cxgb3i/cxgb3i.c b/kernel/drivers/scsi/cxgbi/cxgb3i/cxgb3i.c
new file mode 100644 (file)
index 0000000..3db4c63
--- /dev/null
@@ -0,0 +1,1414 @@
+/*
+ * cxgb3i_offload.c: Chelsio S3xx iscsi offloaded tcp connection management
+ *
+ * Copyright (C) 2003-2008 Chelsio Communications.  All rights reserved.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the LICENSE file included in this
+ * release for licensing terms and conditions.
+ *
+ * Written by: Dimitris Michailidis (dm@chelsio.com)
+ *             Karen Xie (kxie@chelsio.com)
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ":%s: " fmt, __func__
+
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <scsi/scsi_host.h>
+
+#include "common.h"
+#include "t3_cpl.h"
+#include "t3cdev.h"
+#include "cxgb3_defs.h"
+#include "cxgb3_ctl_defs.h"
+#include "cxgb3_offload.h"
+#include "firmware_exports.h"
+#include "cxgb3i.h"
+
+static unsigned int dbg_level;
+#include "../libcxgbi.h"
+
+#define DRV_MODULE_NAME         "cxgb3i"
+#define DRV_MODULE_DESC         "Chelsio T3 iSCSI Driver"
+#define DRV_MODULE_VERSION     "2.0.0"
+#define DRV_MODULE_RELDATE     "Jun. 2010"
+
+static char version[] =
+       DRV_MODULE_DESC " " DRV_MODULE_NAME
+       " v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
+
+MODULE_AUTHOR("Chelsio Communications, Inc.");
+MODULE_DESCRIPTION(DRV_MODULE_DESC);
+MODULE_VERSION(DRV_MODULE_VERSION);
+MODULE_LICENSE("GPL");
+
+module_param(dbg_level, uint, 0644);
+MODULE_PARM_DESC(dbg_level, "debug flag (default=0)");
+
+static int cxgb3i_rcv_win = 256 * 1024;
+module_param(cxgb3i_rcv_win, int, 0644);
+MODULE_PARM_DESC(cxgb3i_rcv_win, "TCP receive window in bytes (default=256KB)");
+
+static int cxgb3i_snd_win = 128 * 1024;
+module_param(cxgb3i_snd_win, int, 0644);
+MODULE_PARM_DESC(cxgb3i_snd_win, "TCP send window in bytes (default=128KB)");
+
+static int cxgb3i_rx_credit_thres = 10 * 1024;
+module_param(cxgb3i_rx_credit_thres, int, 0644);
+MODULE_PARM_DESC(rx_credit_thres,
+                "RX credits return threshold in bytes (default=10KB)");
+
+static unsigned int cxgb3i_max_connect = 8 * 1024;
+module_param(cxgb3i_max_connect, uint, 0644);
+MODULE_PARM_DESC(cxgb3i_max_connect, "Max. # of connections (default=8092)");
+
+static unsigned int cxgb3i_sport_base = 20000;
+module_param(cxgb3i_sport_base, uint, 0644);
+MODULE_PARM_DESC(cxgb3i_sport_base, "starting port number (default=20000)");
+
+static void cxgb3i_dev_open(struct t3cdev *);
+static void cxgb3i_dev_close(struct t3cdev *);
+static void cxgb3i_dev_event_handler(struct t3cdev *, u32, u32);
+
+static struct cxgb3_client t3_client = {
+       .name = DRV_MODULE_NAME,
+       .handlers = cxgb3i_cpl_handlers,
+       .add = cxgb3i_dev_open,
+       .remove = cxgb3i_dev_close,
+       .event_handler = cxgb3i_dev_event_handler,
+};
+
+static struct scsi_host_template cxgb3i_host_template = {
+       .module         = THIS_MODULE,
+       .name           = DRV_MODULE_NAME,
+       .proc_name      = DRV_MODULE_NAME,
+       .can_queue      = CXGB3I_SCSI_HOST_QDEPTH,
+       .queuecommand   = iscsi_queuecommand,
+       .change_queue_depth = scsi_change_queue_depth,
+       .sg_tablesize   = SG_ALL,
+       .max_sectors    = 0xFFFF,
+       .cmd_per_lun    = ISCSI_DEF_CMD_PER_LUN,
+       .eh_abort_handler = iscsi_eh_abort,
+       .eh_device_reset_handler = iscsi_eh_device_reset,
+       .eh_target_reset_handler = iscsi_eh_recover_target,
+       .target_alloc   = iscsi_target_alloc,
+       .use_clustering = DISABLE_CLUSTERING,
+       .this_id        = -1,
+       .track_queue_depth = 1,
+};
+
+static struct iscsi_transport cxgb3i_iscsi_transport = {
+       .owner          = THIS_MODULE,
+       .name           = DRV_MODULE_NAME,
+       /* owner and name should be set already */
+       .caps           = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
+                               | CAP_DATADGST | CAP_DIGEST_OFFLOAD |
+                               CAP_PADDING_OFFLOAD | CAP_TEXT_NEGO,
+       .attr_is_visible        = cxgbi_attr_is_visible,
+       .get_host_param = cxgbi_get_host_param,
+       .set_host_param = cxgbi_set_host_param,
+       /* session management */
+       .create_session = cxgbi_create_session,
+       .destroy_session        = cxgbi_destroy_session,
+       .get_session_param = iscsi_session_get_param,
+       /* connection management */
+       .create_conn    = cxgbi_create_conn,
+       .bind_conn      = cxgbi_bind_conn,
+       .destroy_conn   = iscsi_tcp_conn_teardown,
+       .start_conn     = iscsi_conn_start,
+       .stop_conn      = iscsi_conn_stop,
+       .get_conn_param = iscsi_conn_get_param,
+       .set_param      = cxgbi_set_conn_param,
+       .get_stats      = cxgbi_get_conn_stats,
+       /* pdu xmit req from user space */
+       .send_pdu       = iscsi_conn_send_pdu,
+       /* task */
+       .init_task      = iscsi_tcp_task_init,
+       .xmit_task      = iscsi_tcp_task_xmit,
+       .cleanup_task   = cxgbi_cleanup_task,
+       /* pdu */
+       .alloc_pdu      = cxgbi_conn_alloc_pdu,
+       .init_pdu       = cxgbi_conn_init_pdu,
+       .xmit_pdu       = cxgbi_conn_xmit_pdu,
+       .parse_pdu_itt  = cxgbi_parse_pdu_itt,
+       /* TCP connect/disconnect */
+       .get_ep_param   = cxgbi_get_ep_param,
+       .ep_connect     = cxgbi_ep_connect,
+       .ep_poll        = cxgbi_ep_poll,
+       .ep_disconnect  = cxgbi_ep_disconnect,
+       /* Error recovery timeout call */
+       .session_recovery_timedout = iscsi_session_recovery_timedout,
+};
+
+static struct scsi_transport_template *cxgb3i_stt;
+
+/*
+ * CPL (Chelsio Protocol Language) defines a message passing interface between
+ * the host driver and Chelsio asic.
+ * The section below implments CPLs that related to iscsi tcp connection
+ * open/close/abort and data send/receive.
+ */
+
+static int push_tx_frames(struct cxgbi_sock *csk, int req_completion);
+
+static void send_act_open_req(struct cxgbi_sock *csk, struct sk_buff *skb,
+                             const struct l2t_entry *e)
+{
+       unsigned int wscale = cxgbi_sock_compute_wscale(cxgb3i_rcv_win);
+       struct cpl_act_open_req *req = (struct cpl_act_open_req *)skb->head;
+
+       skb->priority = CPL_PRIORITY_SETUP;
+
+       req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
+       OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, csk->atid));
+       req->local_port = csk->saddr.sin_port;
+       req->peer_port = csk->daddr.sin_port;
+       req->local_ip = csk->saddr.sin_addr.s_addr;
+       req->peer_ip = csk->daddr.sin_addr.s_addr;
+
+       req->opt0h = htonl(V_KEEP_ALIVE(1) | F_TCAM_BYPASS |
+                       V_WND_SCALE(wscale) | V_MSS_IDX(csk->mss_idx) |
+                       V_L2T_IDX(e->idx) | V_TX_CHANNEL(e->smt_idx));
+       req->opt0l = htonl(V_ULP_MODE(ULP2_MODE_ISCSI) |
+                       V_RCV_BUFSIZ(cxgb3i_rcv_win>>10));
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p,%u,0x%lx,%u, %pI4:%u-%pI4:%u, %u,%u,%u.\n",
+               csk, csk->state, csk->flags, csk->atid,
+               &req->local_ip, ntohs(req->local_port),
+               &req->peer_ip, ntohs(req->peer_port),
+               csk->mss_idx, e->idx, e->smt_idx);
+
+       l2t_send(csk->cdev->lldev, skb, csk->l2t);
+}
+
+static inline void act_open_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
+{
+       cxgbi_sock_act_open_req_arp_failure(NULL, skb);
+}
+
+/*
+ * CPL connection close request: host ->
+ *
+ * Close a connection by sending a CPL_CLOSE_CON_REQ message and queue it to
+ * the write queue (i.e., after any unsent txt data).
+ */
+static void send_close_req(struct cxgbi_sock *csk)
+{
+       struct sk_buff *skb = csk->cpl_close;
+       struct cpl_close_con_req *req = (struct cpl_close_con_req *)skb->head;
+       unsigned int tid = csk->tid;
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p,%u,0x%lx,%u.\n",
+               csk, csk->state, csk->flags, csk->tid);
+
+       csk->cpl_close = NULL;
+       req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
+       req->wr.wr_lo = htonl(V_WR_TID(tid));
+       OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, tid));
+       req->rsvd = htonl(csk->write_seq);
+
+       cxgbi_sock_skb_entail(csk, skb);
+       if (csk->state >= CTP_ESTABLISHED)
+               push_tx_frames(csk, 1);
+}
+
+/*
+ * CPL connection abort request: host ->
+ *
+ * Send an ABORT_REQ message. Makes sure we do not send multiple ABORT_REQs
+ * for the same connection and also that we do not try to send a message
+ * after the connection has closed.
+ */
+static void abort_arp_failure(struct t3cdev *tdev, struct sk_buff *skb)
+{
+       struct cpl_abort_req *req = cplhdr(skb);
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "t3dev 0x%p, tid %u, skb 0x%p.\n",
+               tdev, GET_TID(req), skb);
+       req->cmd = CPL_ABORT_NO_RST;
+       cxgb3_ofld_send(tdev, skb);
+}
+
+static void send_abort_req(struct cxgbi_sock *csk)
+{
+       struct sk_buff *skb = csk->cpl_abort_req;
+       struct cpl_abort_req *req;
+
+       if (unlikely(csk->state == CTP_ABORTING || !skb))
+               return;
+       cxgbi_sock_set_state(csk, CTP_ABORTING);
+       cxgbi_sock_set_flag(csk, CTPF_ABORT_RPL_PENDING);
+       /* Purge the send queue so we don't send anything after an abort. */
+       cxgbi_sock_purge_write_queue(csk);
+
+       csk->cpl_abort_req = NULL;
+       req = (struct cpl_abort_req *)skb->head;
+       skb->priority = CPL_PRIORITY_DATA;
+       set_arp_failure_handler(skb, abort_arp_failure);
+       req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
+       req->wr.wr_lo = htonl(V_WR_TID(csk->tid));
+       OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, csk->tid));
+       req->rsvd0 = htonl(csk->snd_nxt);
+       req->rsvd1 = !cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT);
+       req->cmd = CPL_ABORT_SEND_RST;
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p,%u,0x%lx,%u, snd_nxt %u, 0x%x.\n",
+               csk, csk->state, csk->flags, csk->tid, csk->snd_nxt,
+               req->rsvd1);
+
+       l2t_send(csk->cdev->lldev, skb, csk->l2t);
+}
+
+/*
+ * CPL connection abort reply: host ->
+ *
+ * Send an ABORT_RPL message in response of the ABORT_REQ received.
+ */
+static void send_abort_rpl(struct cxgbi_sock *csk, int rst_status)
+{
+       struct sk_buff *skb = csk->cpl_abort_rpl;
+       struct cpl_abort_rpl *rpl = (struct cpl_abort_rpl *)skb->head;
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p,%u,0x%lx,%u, status %d.\n",
+               csk, csk->state, csk->flags, csk->tid, rst_status);
+
+       csk->cpl_abort_rpl = NULL;
+       skb->priority = CPL_PRIORITY_DATA;
+       rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
+       rpl->wr.wr_lo = htonl(V_WR_TID(csk->tid));
+       OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, csk->tid));
+       rpl->cmd = rst_status;
+       cxgb3_ofld_send(csk->cdev->lldev, skb);
+}
+
+/*
+ * CPL connection rx data ack: host ->
+ * Send RX credits through an RX_DATA_ACK CPL message. Returns the number of
+ * credits sent.
+ */
+static u32 send_rx_credits(struct cxgbi_sock *csk, u32 credits)
+{
+       struct sk_buff *skb;
+       struct cpl_rx_data_ack *req;
+       u32 dack = F_RX_DACK_CHANGE | V_RX_DACK_MODE(1);
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
+               "csk 0x%p,%u,0x%lx,%u, credit %u, dack %u.\n",
+               csk, csk->state, csk->flags, csk->tid, credits, dack);
+
+       skb = alloc_wr(sizeof(*req), 0, GFP_ATOMIC);
+       if (!skb) {
+               pr_info("csk 0x%p, credit %u, OOM.\n", csk, credits);
+               return 0;
+       }
+       req = (struct cpl_rx_data_ack *)skb->head;
+       req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
+       OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, csk->tid));
+       req->credit_dack = htonl(F_RX_DACK_CHANGE | V_RX_DACK_MODE(1) |
+                               V_RX_CREDITS(credits));
+       skb->priority = CPL_PRIORITY_ACK;
+       cxgb3_ofld_send(csk->cdev->lldev, skb);
+       return credits;
+}
+
+/*
+ * CPL connection tx data: host ->
+ *
+ * Send iscsi PDU via TX_DATA CPL message. Returns the number of
+ * credits sent.
+ * Each TX_DATA consumes work request credit (wrs), so we need to keep track of
+ * how many we've used so far and how many are pending (i.e., yet ack'ed by T3).
+ */
+
+static unsigned int wrlen __read_mostly;
+static unsigned int skb_wrs[SKB_WR_LIST_SIZE] __read_mostly;
+
+static void init_wr_tab(unsigned int wr_len)
+{
+       int i;
+
+       if (skb_wrs[1])         /* already initialized */
+               return;
+       for (i = 1; i < SKB_WR_LIST_SIZE; i++) {
+               int sgl_len = (3 * i) / 2 + (i & 1);
+
+               sgl_len += 3;
+               skb_wrs[i] = (sgl_len <= wr_len
+                             ? 1 : 1 + (sgl_len - 2) / (wr_len - 1));
+       }
+       wrlen = wr_len * 8;
+}
+
+static inline void make_tx_data_wr(struct cxgbi_sock *csk, struct sk_buff *skb,
+                                  int len, int req_completion)
+{
+       struct tx_data_wr *req;
+       struct l2t_entry *l2t = csk->l2t;
+
+       skb_reset_transport_header(skb);
+       req = (struct tx_data_wr *)__skb_push(skb, sizeof(*req));
+       req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA) |
+                       (req_completion ? F_WR_COMPL : 0));
+       req->wr_lo = htonl(V_WR_TID(csk->tid));
+       /* len includes the length of any HW ULP additions */
+       req->len = htonl(len);
+       /* V_TX_ULP_SUBMODE sets both the mode and submode */
+       req->flags = htonl(V_TX_ULP_SUBMODE(cxgbi_skcb_ulp_mode(skb)) |
+                          V_TX_SHOVE((skb_peek(&csk->write_queue) ? 0 : 1)));
+       req->sndseq = htonl(csk->snd_nxt);
+       req->param = htonl(V_TX_PORT(l2t->smt_idx));
+
+       if (!cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT)) {
+               req->flags |= htonl(V_TX_ACK_PAGES(2) | F_TX_INIT |
+                                   V_TX_CPU_IDX(csk->rss_qid));
+               /* sendbuffer is in units of 32KB. */
+               req->param |= htonl(V_TX_SNDBUF(cxgb3i_snd_win >> 15));
+               cxgbi_sock_set_flag(csk, CTPF_TX_DATA_SENT);
+       }
+}
+
+/**
+ * push_tx_frames -- start transmit
+ * @c3cn: the offloaded connection
+ * @req_completion: request wr_ack or not
+ *
+ * Prepends TX_DATA_WR or CPL_CLOSE_CON_REQ headers to buffers waiting in a
+ * connection's send queue and sends them on to T3.  Must be called with the
+ * connection's lock held.  Returns the amount of send buffer space that was
+ * freed as a result of sending queued data to T3.
+ */
+
+static void arp_failure_skb_discard(struct t3cdev *dev, struct sk_buff *skb)
+{
+       kfree_skb(skb);
+}
+
+static int push_tx_frames(struct cxgbi_sock *csk, int req_completion)
+{
+       int total_size = 0;
+       struct sk_buff *skb;
+
+       if (unlikely(csk->state < CTP_ESTABLISHED ||
+               csk->state == CTP_CLOSE_WAIT_1 || csk->state >= CTP_ABORTING)) {
+                       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_TX,
+                               "csk 0x%p,%u,0x%lx,%u, in closing state.\n",
+                               csk, csk->state, csk->flags, csk->tid);
+               return 0;
+       }
+
+       while (csk->wr_cred && (skb = skb_peek(&csk->write_queue)) != NULL) {
+               int len = skb->len;     /* length before skb_push */
+               int frags = skb_shinfo(skb)->nr_frags + (len != skb->data_len);
+               int wrs_needed = skb_wrs[frags];
+
+               if (wrs_needed > 1 && len + sizeof(struct tx_data_wr) <= wrlen)
+                       wrs_needed = 1;
+
+               WARN_ON(frags >= SKB_WR_LIST_SIZE || wrs_needed < 1);
+
+               if (csk->wr_cred < wrs_needed) {
+                       log_debug(1 << CXGBI_DBG_PDU_TX,
+                               "csk 0x%p, skb len %u/%u, frag %u, wr %d<%u.\n",
+                               csk, skb->len, skb->data_len, frags,
+                               wrs_needed, csk->wr_cred);
+                       break;
+               }
+
+               __skb_unlink(skb, &csk->write_queue);
+               skb->priority = CPL_PRIORITY_DATA;
+               skb->csum = wrs_needed; /* remember this until the WR_ACK */
+               csk->wr_cred -= wrs_needed;
+               csk->wr_una_cred += wrs_needed;
+               cxgbi_sock_enqueue_wr(csk, skb);
+
+               log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_TX,
+                       "csk 0x%p, enqueue, skb len %u/%u, frag %u, wr %d, "
+                       "left %u, unack %u.\n",
+                       csk, skb->len, skb->data_len, frags, skb->csum,
+                       csk->wr_cred, csk->wr_una_cred);
+
+               if (likely(cxgbi_skcb_test_flag(skb, SKCBF_TX_NEED_HDR))) {
+                       if ((req_completion &&
+                               csk->wr_una_cred == wrs_needed) ||
+                            csk->wr_una_cred >= csk->wr_max_cred / 2) {
+                               req_completion = 1;
+                               csk->wr_una_cred = 0;
+                       }
+                       len += cxgbi_ulp_extra_len(cxgbi_skcb_ulp_mode(skb));
+                       make_tx_data_wr(csk, skb, len, req_completion);
+                       csk->snd_nxt += len;
+                       cxgbi_skcb_clear_flag(skb, SKCBF_TX_NEED_HDR);
+               }
+               total_size += skb->truesize;
+               log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_TX,
+                       "csk 0x%p, tid 0x%x, send skb 0x%p.\n",
+                       csk, csk->tid, skb);
+               set_arp_failure_handler(skb, arp_failure_skb_discard);
+               l2t_send(csk->cdev->lldev, skb, csk->l2t);
+       }
+       return total_size;
+}
+
+/*
+ * Process a CPL_ACT_ESTABLISH message: -> host
+ * Updates connection state from an active establish CPL message.  Runs with
+ * the connection lock held.
+ */
+
+static inline void free_atid(struct cxgbi_sock *csk)
+{
+       if (cxgbi_sock_flag(csk, CTPF_HAS_ATID)) {
+               cxgb3_free_atid(csk->cdev->lldev, csk->atid);
+               cxgbi_sock_clear_flag(csk, CTPF_HAS_ATID);
+               cxgbi_sock_put(csk);
+       }
+}
+
+static int do_act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
+{
+       struct cxgbi_sock *csk = ctx;
+       struct cpl_act_establish *req = cplhdr(skb);
+       unsigned int tid = GET_TID(req);
+       unsigned int atid = G_PASS_OPEN_TID(ntohl(req->tos_tid));
+       u32 rcv_isn = ntohl(req->rcv_isn);      /* real RCV_ISN + 1 */
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "atid 0x%x,tid 0x%x, csk 0x%p,%u,0x%lx, isn %u.\n",
+               atid, atid, csk, csk->state, csk->flags, rcv_isn);
+
+       cxgbi_sock_get(csk);
+       cxgbi_sock_set_flag(csk, CTPF_HAS_TID);
+       csk->tid = tid;
+       cxgb3_insert_tid(csk->cdev->lldev, &t3_client, csk, tid);
+
+       free_atid(csk);
+
+       csk->rss_qid = G_QNUM(ntohs(skb->csum));
+
+       spin_lock_bh(&csk->lock);
+       if (csk->retry_timer.function) {
+               del_timer(&csk->retry_timer);
+               csk->retry_timer.function = NULL;
+       }
+
+       if (unlikely(csk->state != CTP_ACTIVE_OPEN))
+               pr_info("csk 0x%p,%u,0x%lx,%u, got EST.\n",
+                       csk, csk->state, csk->flags, csk->tid);
+
+       csk->copied_seq = csk->rcv_wup = csk->rcv_nxt = rcv_isn;
+       if (cxgb3i_rcv_win > (M_RCV_BUFSIZ << 10))
+               csk->rcv_wup -= cxgb3i_rcv_win - (M_RCV_BUFSIZ << 10);
+
+       cxgbi_sock_established(csk, ntohl(req->snd_isn), ntohs(req->tcp_opt));
+
+       if (unlikely(cxgbi_sock_flag(csk, CTPF_ACTIVE_CLOSE_NEEDED)))
+               /* upper layer has requested closing */
+               send_abort_req(csk);
+       else {
+               if (skb_queue_len(&csk->write_queue))
+                       push_tx_frames(csk, 1);
+               cxgbi_conn_tx_open(csk);
+       }
+
+       spin_unlock_bh(&csk->lock);
+       __kfree_skb(skb);
+       return 0;
+}
+
+/*
+ * Process a CPL_ACT_OPEN_RPL message: -> host
+ * Handle active open failures.
+ */
+static int act_open_rpl_status_to_errno(int status)
+{
+       switch (status) {
+       case CPL_ERR_CONN_RESET:
+               return -ECONNREFUSED;
+       case CPL_ERR_ARP_MISS:
+               return -EHOSTUNREACH;
+       case CPL_ERR_CONN_TIMEDOUT:
+               return -ETIMEDOUT;
+       case CPL_ERR_TCAM_FULL:
+               return -ENOMEM;
+       case CPL_ERR_CONN_EXIST:
+               return -EADDRINUSE;
+       default:
+               return -EIO;
+       }
+}
+
+static void act_open_retry_timer(unsigned long data)
+{
+       struct sk_buff *skb;
+       struct cxgbi_sock *csk = (struct cxgbi_sock *)data;
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p,%u,0x%lx,%u.\n",
+               csk, csk->state, csk->flags, csk->tid);
+
+       cxgbi_sock_get(csk);
+       spin_lock_bh(&csk->lock);
+       skb = alloc_wr(sizeof(struct cpl_act_open_req), 0, GFP_ATOMIC);
+       if (!skb)
+               cxgbi_sock_fail_act_open(csk, -ENOMEM);
+       else {
+               skb->sk = (struct sock *)csk;
+               set_arp_failure_handler(skb, act_open_arp_failure);
+               send_act_open_req(csk, skb, csk->l2t);
+       }
+       spin_unlock_bh(&csk->lock);
+       cxgbi_sock_put(csk);
+}
+
+static int do_act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
+{
+       struct cxgbi_sock *csk = ctx;
+       struct cpl_act_open_rpl *rpl = cplhdr(skb);
+
+       pr_info("csk 0x%p,%u,0x%lx,%u, status %u, %pI4:%u-%pI4:%u.\n",
+               csk, csk->state, csk->flags, csk->atid, rpl->status,
+               &csk->saddr.sin_addr.s_addr, ntohs(csk->saddr.sin_port),
+               &csk->daddr.sin_addr.s_addr, ntohs(csk->daddr.sin_port));
+
+       if (rpl->status != CPL_ERR_TCAM_FULL &&
+           rpl->status != CPL_ERR_CONN_EXIST &&
+           rpl->status != CPL_ERR_ARP_MISS)
+               cxgb3_queue_tid_release(tdev, GET_TID(rpl));
+
+       cxgbi_sock_get(csk);
+       spin_lock_bh(&csk->lock);
+       if (rpl->status == CPL_ERR_CONN_EXIST &&
+           csk->retry_timer.function != act_open_retry_timer) {
+               csk->retry_timer.function = act_open_retry_timer;
+               mod_timer(&csk->retry_timer, jiffies + HZ / 2);
+       } else
+               cxgbi_sock_fail_act_open(csk,
+                               act_open_rpl_status_to_errno(rpl->status));
+
+       spin_unlock_bh(&csk->lock);
+       cxgbi_sock_put(csk);
+       __kfree_skb(skb);
+       return 0;
+}
+
+/*
+ * Process PEER_CLOSE CPL messages: -> host
+ * Handle peer FIN.
+ */
+static int do_peer_close(struct t3cdev *cdev, struct sk_buff *skb, void *ctx)
+{
+       struct cxgbi_sock *csk = ctx;
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p,%u,0x%lx,%u.\n",
+               csk, csk->state, csk->flags, csk->tid);
+
+       cxgbi_sock_rcv_peer_close(csk);
+       __kfree_skb(skb);
+       return 0;
+}
+
+/*
+ * Process CLOSE_CONN_RPL CPL message: -> host
+ * Process a peer ACK to our FIN.
+ */
+static int do_close_con_rpl(struct t3cdev *cdev, struct sk_buff *skb,
+                           void *ctx)
+{
+       struct cxgbi_sock *csk = ctx;
+       struct cpl_close_con_rpl *rpl = cplhdr(skb);
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p,%u,0x%lx,%u, snxt %u.\n",
+               csk, csk->state, csk->flags, csk->tid, ntohl(rpl->snd_nxt));
+
+       cxgbi_sock_rcv_close_conn_rpl(csk, ntohl(rpl->snd_nxt));
+       __kfree_skb(skb);
+       return 0;
+}
+
+/*
+ * Process ABORT_REQ_RSS CPL message: -> host
+ * Process abort requests.  If we are waiting for an ABORT_RPL we ignore this
+ * request except that we need to reply to it.
+ */
+
+static int abort_status_to_errno(struct cxgbi_sock *csk, int abort_reason,
+                                int *need_rst)
+{
+       switch (abort_reason) {
+       case CPL_ERR_BAD_SYN: /* fall through */
+       case CPL_ERR_CONN_RESET:
+               return csk->state > CTP_ESTABLISHED ? -EPIPE : -ECONNRESET;
+       case CPL_ERR_XMIT_TIMEDOUT:
+       case CPL_ERR_PERSIST_TIMEDOUT:
+       case CPL_ERR_FINWAIT2_TIMEDOUT:
+       case CPL_ERR_KEEPALIVE_TIMEDOUT:
+               return -ETIMEDOUT;
+       default:
+               return -EIO;
+       }
+}
+
+static int do_abort_req(struct t3cdev *cdev, struct sk_buff *skb, void *ctx)
+{
+       const struct cpl_abort_req_rss *req = cplhdr(skb);
+       struct cxgbi_sock *csk = ctx;
+       int rst_status = CPL_ABORT_NO_RST;
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p,%u,0x%lx,%u.\n",
+               csk, csk->state, csk->flags, csk->tid);
+
+       if (req->status == CPL_ERR_RTX_NEG_ADVICE ||
+           req->status == CPL_ERR_PERSIST_NEG_ADVICE) {
+               goto done;
+       }
+
+       cxgbi_sock_get(csk);
+       spin_lock_bh(&csk->lock);
+
+       if (!cxgbi_sock_flag(csk, CTPF_ABORT_REQ_RCVD)) {
+               cxgbi_sock_set_flag(csk, CTPF_ABORT_REQ_RCVD);
+               cxgbi_sock_set_state(csk, CTP_ABORTING);
+               goto out;
+       }
+
+       cxgbi_sock_clear_flag(csk, CTPF_ABORT_REQ_RCVD);
+       send_abort_rpl(csk, rst_status);
+
+       if (!cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING)) {
+               csk->err = abort_status_to_errno(csk, req->status, &rst_status);
+               cxgbi_sock_closed(csk);
+       }
+
+out:
+       spin_unlock_bh(&csk->lock);
+       cxgbi_sock_put(csk);
+done:
+       __kfree_skb(skb);
+       return 0;
+}
+
+/*
+ * Process ABORT_RPL_RSS CPL message: -> host
+ * Process abort replies.  We only process these messages if we anticipate
+ * them as the coordination between SW and HW in this area is somewhat lacking
+ * and sometimes we get ABORT_RPLs after we are done with the connection that
+ * originated the ABORT_REQ.
+ */
+static int do_abort_rpl(struct t3cdev *cdev, struct sk_buff *skb, void *ctx)
+{
+       struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
+       struct cxgbi_sock *csk = ctx;
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "status 0x%x, csk 0x%p, s %u, 0x%lx.\n",
+               rpl->status, csk, csk ? csk->state : 0,
+               csk ? csk->flags : 0UL);
+       /*
+        * Ignore replies to post-close aborts indicating that the abort was
+        * requested too late.  These connections are terminated when we get
+        * PEER_CLOSE or CLOSE_CON_RPL and by the time the abort_rpl_rss
+        * arrives the TID is either no longer used or it has been recycled.
+        */
+       if (rpl->status == CPL_ERR_ABORT_FAILED)
+               goto rel_skb;
+       /*
+        * Sometimes we've already closed the connection, e.g., a post-close
+        * abort races with ABORT_REQ_RSS, the latter frees the connection
+        * expecting the ABORT_REQ will fail with CPL_ERR_ABORT_FAILED,
+        * but FW turns the ABORT_REQ into a regular one and so we get
+        * ABORT_RPL_RSS with status 0 and no connection.
+        */
+       if (csk)
+               cxgbi_sock_rcv_abort_rpl(csk);
+rel_skb:
+       __kfree_skb(skb);
+       return 0;
+}
+
+/*
+ * Process RX_ISCSI_HDR CPL message: -> host
+ * Handle received PDUs, the payload could be DDP'ed. If not, the payload
+ * follow after the bhs.
+ */
+static int do_iscsi_hdr(struct t3cdev *t3dev, struct sk_buff *skb, void *ctx)
+{
+       struct cxgbi_sock *csk = ctx;
+       struct cpl_iscsi_hdr *hdr_cpl = cplhdr(skb);
+       struct cpl_iscsi_hdr_norss data_cpl;
+       struct cpl_rx_data_ddp_norss ddp_cpl;
+       unsigned int hdr_len, data_len, status;
+       unsigned int len;
+       int err;
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
+               "csk 0x%p,%u,0x%lx,%u, skb 0x%p,%u.\n",
+               csk, csk->state, csk->flags, csk->tid, skb, skb->len);
+
+       spin_lock_bh(&csk->lock);
+
+       if (unlikely(csk->state >= CTP_PASSIVE_CLOSE)) {
+               log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+                       "csk 0x%p,%u,0x%lx,%u, bad state.\n",
+                       csk, csk->state, csk->flags, csk->tid);
+               if (csk->state != CTP_ABORTING)
+                       goto abort_conn;
+               else
+                       goto discard;
+       }
+
+       cxgbi_skcb_tcp_seq(skb) = ntohl(hdr_cpl->seq);
+       cxgbi_skcb_flags(skb) = 0;
+
+       skb_reset_transport_header(skb);
+       __skb_pull(skb, sizeof(struct cpl_iscsi_hdr));
+
+       len = hdr_len = ntohs(hdr_cpl->len);
+       /* msg coalesce is off or not enough data received */
+       if (skb->len <= hdr_len) {
+               pr_err("%s: tid %u, CPL_ISCSI_HDR, skb len %u < %u.\n",
+                       csk->cdev->ports[csk->port_id]->name, csk->tid,
+                       skb->len, hdr_len);
+               goto abort_conn;
+       }
+       cxgbi_skcb_set_flag(skb, SKCBF_RX_COALESCED);
+
+       err = skb_copy_bits(skb, skb->len - sizeof(ddp_cpl), &ddp_cpl,
+                           sizeof(ddp_cpl));
+       if (err < 0) {
+               pr_err("%s: tid %u, copy cpl_ddp %u-%zu failed %d.\n",
+                       csk->cdev->ports[csk->port_id]->name, csk->tid,
+                       skb->len, sizeof(ddp_cpl), err);
+               goto abort_conn;
+       }
+
+       cxgbi_skcb_set_flag(skb, SKCBF_RX_STATUS);
+       cxgbi_skcb_rx_pdulen(skb) = ntohs(ddp_cpl.len);
+       cxgbi_skcb_rx_ddigest(skb) = ntohl(ddp_cpl.ulp_crc);
+       status = ntohl(ddp_cpl.ddp_status);
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
+               "csk 0x%p, skb 0x%p,%u, pdulen %u, status 0x%x.\n",
+               csk, skb, skb->len, cxgbi_skcb_rx_pdulen(skb), status);
+
+       if (status & (1 << CPL_RX_DDP_STATUS_HCRC_SHIFT))
+               cxgbi_skcb_set_flag(skb, SKCBF_RX_HCRC_ERR);
+       if (status & (1 << CPL_RX_DDP_STATUS_DCRC_SHIFT))
+               cxgbi_skcb_set_flag(skb, SKCBF_RX_DCRC_ERR);
+       if (status & (1 << CPL_RX_DDP_STATUS_PAD_SHIFT))
+               cxgbi_skcb_set_flag(skb, SKCBF_RX_PAD_ERR);
+
+       if (skb->len > (hdr_len + sizeof(ddp_cpl))) {
+               err = skb_copy_bits(skb, hdr_len, &data_cpl, sizeof(data_cpl));
+               if (err < 0) {
+                       pr_err("%s: tid %u, cp %zu/%u failed %d.\n",
+                               csk->cdev->ports[csk->port_id]->name,
+                               csk->tid, sizeof(data_cpl), skb->len, err);
+                       goto abort_conn;
+               }
+               data_len = ntohs(data_cpl.len);
+               log_debug(1 << CXGBI_DBG_DDP | 1 << CXGBI_DBG_PDU_RX,
+                       "skb 0x%p, pdu not ddp'ed %u/%u, status 0x%x.\n",
+                       skb, data_len, cxgbi_skcb_rx_pdulen(skb), status);
+               len += sizeof(data_cpl) + data_len;
+       } else if (status & (1 << CPL_RX_DDP_STATUS_DDP_SHIFT))
+               cxgbi_skcb_set_flag(skb, SKCBF_RX_DATA_DDPD);
+
+       csk->rcv_nxt = ntohl(ddp_cpl.seq) + cxgbi_skcb_rx_pdulen(skb);
+       __pskb_trim(skb, len);
+       __skb_queue_tail(&csk->receive_queue, skb);
+       cxgbi_conn_pdu_ready(csk);
+
+       spin_unlock_bh(&csk->lock);
+       return 0;
+
+abort_conn:
+       send_abort_req(csk);
+discard:
+       spin_unlock_bh(&csk->lock);
+       __kfree_skb(skb);
+       return 0;
+}
+
+/*
+ * Process TX_DATA_ACK CPL messages: -> host
+ * Process an acknowledgment of WR completion.  Advance snd_una and send the
+ * next batch of work requests from the write queue.
+ */
+static int do_wr_ack(struct t3cdev *cdev, struct sk_buff *skb, void *ctx)
+{
+       struct cxgbi_sock *csk = ctx;
+       struct cpl_wr_ack *hdr = cplhdr(skb);
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
+               "csk 0x%p,%u,0x%lx,%u, cr %u.\n",
+               csk, csk->state, csk->flags, csk->tid, ntohs(hdr->credits));
+
+       cxgbi_sock_rcv_wr_ack(csk, ntohs(hdr->credits), ntohl(hdr->snd_una), 1);
+       __kfree_skb(skb);
+       return 0;
+}
+
+/*
+ * for each connection, pre-allocate skbs needed for close/abort requests. So
+ * that we can service the request right away.
+ */
+static int alloc_cpls(struct cxgbi_sock *csk)
+{
+       csk->cpl_close = alloc_wr(sizeof(struct cpl_close_con_req), 0,
+                                       GFP_KERNEL);
+       if (!csk->cpl_close)
+               return -ENOMEM;
+       csk->cpl_abort_req = alloc_wr(sizeof(struct cpl_abort_req), 0,
+                                       GFP_KERNEL);
+       if (!csk->cpl_abort_req)
+               goto free_cpl_skbs;
+
+       csk->cpl_abort_rpl = alloc_wr(sizeof(struct cpl_abort_rpl), 0,
+                                       GFP_KERNEL);
+       if (!csk->cpl_abort_rpl)
+               goto free_cpl_skbs;
+
+       return 0;
+
+free_cpl_skbs:
+       cxgbi_sock_free_cpl_skbs(csk);
+       return -ENOMEM;
+}
+
+/**
+ * release_offload_resources - release offload resource
+ * @c3cn: the offloaded iscsi tcp connection.
+ * Release resources held by an offload connection (TID, L2T entry, etc.)
+ */
+static void l2t_put(struct cxgbi_sock *csk)
+{
+       struct t3cdev *t3dev = (struct t3cdev *)csk->cdev->lldev;
+
+       if (csk->l2t) {
+               l2t_release(t3dev, csk->l2t);
+               csk->l2t = NULL;
+               cxgbi_sock_put(csk);
+       }
+}
+
+static void release_offload_resources(struct cxgbi_sock *csk)
+{
+       struct t3cdev *t3dev = (struct t3cdev *)csk->cdev->lldev;
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p,%u,0x%lx,%u.\n",
+               csk, csk->state, csk->flags, csk->tid);
+
+       csk->rss_qid = 0;
+       cxgbi_sock_free_cpl_skbs(csk);
+
+       if (csk->wr_cred != csk->wr_max_cred) {
+               cxgbi_sock_purge_wr_queue(csk);
+               cxgbi_sock_reset_wr_list(csk);
+       }
+       l2t_put(csk);
+       if (cxgbi_sock_flag(csk, CTPF_HAS_ATID))
+               free_atid(csk);
+       else if (cxgbi_sock_flag(csk, CTPF_HAS_TID)) {
+               cxgb3_remove_tid(t3dev, (void *)csk, csk->tid);
+               cxgbi_sock_clear_flag(csk, CTPF_HAS_TID);
+               cxgbi_sock_put(csk);
+       }
+       csk->dst = NULL;
+       csk->cdev = NULL;
+}
+
+static void update_address(struct cxgbi_hba *chba)
+{
+       if (chba->ipv4addr) {
+               if (chba->vdev &&
+                   chba->ipv4addr != cxgb3i_get_private_ipv4addr(chba->vdev)) {
+                       cxgb3i_set_private_ipv4addr(chba->vdev, chba->ipv4addr);
+                       cxgb3i_set_private_ipv4addr(chba->ndev, 0);
+                       pr_info("%s set %pI4.\n",
+                               chba->vdev->name, &chba->ipv4addr);
+               } else if (chba->ipv4addr !=
+                               cxgb3i_get_private_ipv4addr(chba->ndev)) {
+                       cxgb3i_set_private_ipv4addr(chba->ndev, chba->ipv4addr);
+                       pr_info("%s set %pI4.\n",
+                               chba->ndev->name, &chba->ipv4addr);
+               }
+       } else if (cxgb3i_get_private_ipv4addr(chba->ndev)) {
+               if (chba->vdev)
+                       cxgb3i_set_private_ipv4addr(chba->vdev, 0);
+               cxgb3i_set_private_ipv4addr(chba->ndev, 0);
+       }
+}
+
+static int init_act_open(struct cxgbi_sock *csk)
+{
+       struct dst_entry *dst = csk->dst;
+       struct cxgbi_device *cdev = csk->cdev;
+       struct t3cdev *t3dev = (struct t3cdev *)cdev->lldev;
+       struct net_device *ndev = cdev->ports[csk->port_id];
+       struct cxgbi_hba *chba = cdev->hbas[csk->port_id];
+       struct sk_buff *skb = NULL;
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p,%u,0x%lx.\n", csk, csk->state, csk->flags);
+
+       update_address(chba);
+       if (chba->ipv4addr)
+               csk->saddr.sin_addr.s_addr = chba->ipv4addr;
+
+       csk->rss_qid = 0;
+       csk->l2t = t3_l2t_get(t3dev, dst, ndev,
+                             &csk->daddr.sin_addr.s_addr);
+       if (!csk->l2t) {
+               pr_err("NO l2t available.\n");
+               return -EINVAL;
+       }
+       cxgbi_sock_get(csk);
+
+       csk->atid = cxgb3_alloc_atid(t3dev, &t3_client, csk);
+       if (csk->atid < 0) {
+               pr_err("NO atid available.\n");
+               goto rel_resource;
+       }
+       cxgbi_sock_set_flag(csk, CTPF_HAS_ATID);
+       cxgbi_sock_get(csk);
+
+       skb = alloc_wr(sizeof(struct cpl_act_open_req), 0, GFP_KERNEL);
+       if (!skb)
+               goto rel_resource;
+       skb->sk = (struct sock *)csk;
+       set_arp_failure_handler(skb, act_open_arp_failure);
+
+       csk->wr_max_cred = csk->wr_cred = T3C_DATA(t3dev)->max_wrs - 1;
+       csk->wr_una_cred = 0;
+       csk->mss_idx = cxgbi_sock_select_mss(csk, dst_mtu(dst));
+       cxgbi_sock_reset_wr_list(csk);
+       csk->err = 0;
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p,%u,0x%lx, %pI4:%u-%pI4:%u.\n",
+               csk, csk->state, csk->flags,
+               &csk->saddr.sin_addr.s_addr, ntohs(csk->saddr.sin_port),
+               &csk->daddr.sin_addr.s_addr, ntohs(csk->daddr.sin_port));
+
+       cxgbi_sock_set_state(csk, CTP_ACTIVE_OPEN);
+       send_act_open_req(csk, skb, csk->l2t);
+       return 0;
+
+rel_resource:
+       if (skb)
+               __kfree_skb(skb);
+       return -EINVAL;
+}
+
+cxgb3_cpl_handler_func cxgb3i_cpl_handlers[NUM_CPL_CMDS] = {
+       [CPL_ACT_ESTABLISH] = do_act_establish,
+       [CPL_ACT_OPEN_RPL] = do_act_open_rpl,
+       [CPL_PEER_CLOSE] = do_peer_close,
+       [CPL_ABORT_REQ_RSS] = do_abort_req,
+       [CPL_ABORT_RPL_RSS] = do_abort_rpl,
+       [CPL_CLOSE_CON_RPL] = do_close_con_rpl,
+       [CPL_TX_DMA_ACK] = do_wr_ack,
+       [CPL_ISCSI_HDR] = do_iscsi_hdr,
+};
+
+/**
+ * cxgb3i_ofld_init - allocate and initialize resources for each adapter found
+ * @cdev:      cxgbi adapter
+ */
+int cxgb3i_ofld_init(struct cxgbi_device *cdev)
+{
+       struct t3cdev *t3dev = (struct t3cdev *)cdev->lldev;
+       struct adap_ports port;
+       struct ofld_page_info rx_page_info;
+       unsigned int wr_len;
+       int rc;
+
+       if (t3dev->ctl(t3dev, GET_WR_LEN, &wr_len) < 0 ||
+           t3dev->ctl(t3dev, GET_PORTS, &port) < 0 ||
+           t3dev->ctl(t3dev, GET_RX_PAGE_INFO, &rx_page_info) < 0) {
+               pr_warn("t3 0x%p, offload up, ioctl failed.\n", t3dev);
+               return -EINVAL;
+       }
+
+       if (cxgb3i_max_connect > CXGBI_MAX_CONN)
+               cxgb3i_max_connect = CXGBI_MAX_CONN;
+
+       rc = cxgbi_device_portmap_create(cdev, cxgb3i_sport_base,
+                                       cxgb3i_max_connect);
+       if (rc < 0)
+               return rc;
+
+       init_wr_tab(wr_len);
+       cdev->csk_release_offload_resources = release_offload_resources;
+       cdev->csk_push_tx_frames = push_tx_frames;
+       cdev->csk_send_abort_req = send_abort_req;
+       cdev->csk_send_close_req = send_close_req;
+       cdev->csk_send_rx_credits = send_rx_credits;
+       cdev->csk_alloc_cpls = alloc_cpls;
+       cdev->csk_init_act_open = init_act_open;
+
+       pr_info("cdev 0x%p, offload up, added.\n", cdev);
+       return 0;
+}
+
+/*
+ * functions to program the pagepod in h/w
+ */
+static inline void ulp_mem_io_set_hdr(struct sk_buff *skb, unsigned int addr)
+{
+       struct ulp_mem_io *req = (struct ulp_mem_io *)skb->head;
+
+       memset(req, 0, sizeof(*req));
+
+       req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_BYPASS));
+       req->cmd_lock_addr = htonl(V_ULP_MEMIO_ADDR(addr >> 5) |
+                                  V_ULPTX_CMD(ULP_MEM_WRITE));
+       req->len = htonl(V_ULP_MEMIO_DATA_LEN(PPOD_SIZE >> 5) |
+                        V_ULPTX_NFLITS((PPOD_SIZE >> 3) + 1));
+}
+
+static int ddp_set_map(struct cxgbi_sock *csk, struct cxgbi_pagepod_hdr *hdr,
+                       unsigned int idx, unsigned int npods,
+                               struct cxgbi_gather_list *gl)
+{
+       struct cxgbi_device *cdev = csk->cdev;
+       struct cxgbi_ddp_info *ddp = cdev->ddp;
+       unsigned int pm_addr = (idx << PPOD_SIZE_SHIFT) + ddp->llimit;
+       int i;
+
+       log_debug(1 << CXGBI_DBG_DDP,
+               "csk 0x%p, idx %u, npods %u, gl 0x%p.\n",
+               csk, idx, npods, gl);
+
+       for (i = 0; i < npods; i++, idx++, pm_addr += PPOD_SIZE) {
+               struct sk_buff *skb = alloc_wr(sizeof(struct ulp_mem_io) +
+                                               PPOD_SIZE, 0, GFP_ATOMIC);
+
+               if (!skb)
+                       return -ENOMEM;
+
+               ulp_mem_io_set_hdr(skb, pm_addr);
+               cxgbi_ddp_ppod_set((struct cxgbi_pagepod *)(skb->head +
+                                       sizeof(struct ulp_mem_io)),
+                                  hdr, gl, i * PPOD_PAGES_MAX);
+               skb->priority = CPL_PRIORITY_CONTROL;
+               cxgb3_ofld_send(cdev->lldev, skb);
+       }
+       return 0;
+}
+
+static void ddp_clear_map(struct cxgbi_hba *chba, unsigned int tag,
+                         unsigned int idx, unsigned int npods)
+{
+       struct cxgbi_device *cdev = chba->cdev;
+       struct cxgbi_ddp_info *ddp = cdev->ddp;
+       unsigned int pm_addr = (idx << PPOD_SIZE_SHIFT) + ddp->llimit;
+       int i;
+
+       log_debug(1 << CXGBI_DBG_DDP,
+               "cdev 0x%p, idx %u, npods %u, tag 0x%x.\n",
+               cdev, idx, npods, tag);
+
+       for (i = 0; i < npods; i++, idx++, pm_addr += PPOD_SIZE) {
+               struct sk_buff *skb = alloc_wr(sizeof(struct ulp_mem_io) +
+                                               PPOD_SIZE, 0, GFP_ATOMIC);
+
+               if (!skb) {
+                       pr_err("tag 0x%x, 0x%x, %d/%u, skb OOM.\n",
+                               tag, idx, i, npods);
+                       continue;
+               }
+               ulp_mem_io_set_hdr(skb, pm_addr);
+               skb->priority = CPL_PRIORITY_CONTROL;
+               cxgb3_ofld_send(cdev->lldev, skb);
+       }
+}
+
+static int ddp_setup_conn_pgidx(struct cxgbi_sock *csk,
+                                      unsigned int tid, int pg_idx, bool reply)
+{
+       struct sk_buff *skb = alloc_wr(sizeof(struct cpl_set_tcb_field), 0,
+                                       GFP_KERNEL);
+       struct cpl_set_tcb_field *req;
+       u64 val = pg_idx < DDP_PGIDX_MAX ? pg_idx : 0;
+
+       log_debug(1 << CXGBI_DBG_DDP,
+               "csk 0x%p, tid %u, pg_idx %d.\n", csk, tid, pg_idx);
+       if (!skb)
+               return -ENOMEM;
+
+       /* set up ulp submode and page size */
+       req = (struct cpl_set_tcb_field *)skb->head;
+       req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
+       OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
+       req->reply = V_NO_REPLY(reply ? 0 : 1);
+       req->cpu_idx = 0;
+       req->word = htons(31);
+       req->mask = cpu_to_be64(0xF0000000);
+       req->val = cpu_to_be64(val << 28);
+       skb->priority = CPL_PRIORITY_CONTROL;
+
+       cxgb3_ofld_send(csk->cdev->lldev, skb);
+       return 0;
+}
+
+/**
+ * cxgb3i_setup_conn_digest - setup conn. digest setting
+ * @csk: cxgb tcp socket
+ * @tid: connection id
+ * @hcrc: header digest enabled
+ * @dcrc: data digest enabled
+ * @reply: request reply from h/w
+ * set up the iscsi digest settings for a connection identified by tid
+ */
+static int ddp_setup_conn_digest(struct cxgbi_sock *csk, unsigned int tid,
+                            int hcrc, int dcrc, int reply)
+{
+       struct sk_buff *skb = alloc_wr(sizeof(struct cpl_set_tcb_field), 0,
+                                       GFP_KERNEL);
+       struct cpl_set_tcb_field *req;
+       u64 val = (hcrc ? 1 : 0) | (dcrc ? 2 : 0);
+
+       log_debug(1 << CXGBI_DBG_DDP,
+               "csk 0x%p, tid %u, crc %d,%d.\n", csk, tid, hcrc, dcrc);
+       if (!skb)
+               return -ENOMEM;
+
+       /* set up ulp submode and page size */
+       req = (struct cpl_set_tcb_field *)skb->head;
+       req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
+       OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
+       req->reply = V_NO_REPLY(reply ? 0 : 1);
+       req->cpu_idx = 0;
+       req->word = htons(31);
+       req->mask = cpu_to_be64(0x0F000000);
+       req->val = cpu_to_be64(val << 24);
+       skb->priority = CPL_PRIORITY_CONTROL;
+
+       cxgb3_ofld_send(csk->cdev->lldev, skb);
+       return 0;
+}
+
+/**
+ * t3_ddp_cleanup - release the cxgb3 adapter's ddp resource
+ * @cdev: cxgb3i adapter
+ * release all the resource held by the ddp pagepod manager for a given
+ * adapter if needed
+ */
+
+static void t3_ddp_cleanup(struct cxgbi_device *cdev)
+{
+       struct t3cdev *tdev = (struct t3cdev *)cdev->lldev;
+
+       if (cxgbi_ddp_cleanup(cdev)) {
+               pr_info("t3dev 0x%p, ulp_iscsi no more user.\n", tdev);
+               tdev->ulp_iscsi = NULL;
+       }
+}
+
+/**
+ * ddp_init - initialize the cxgb3 adapter's ddp resource
+ * @cdev: cxgb3i adapter
+ * initialize the ddp pagepod manager for a given adapter
+ */
+static int cxgb3i_ddp_init(struct cxgbi_device *cdev)
+{
+       struct t3cdev *tdev = (struct t3cdev *)cdev->lldev;
+       struct cxgbi_ddp_info *ddp = tdev->ulp_iscsi;
+       struct ulp_iscsi_info uinfo;
+       unsigned int pgsz_factor[4];
+       int i, err;
+
+       if (ddp) {
+               kref_get(&ddp->refcnt);
+               pr_warn("t3dev 0x%p, ddp 0x%p already set up.\n",
+                       tdev, tdev->ulp_iscsi);
+               cdev->ddp = ddp;
+               return -EALREADY;
+       }
+
+       err = tdev->ctl(tdev, ULP_ISCSI_GET_PARAMS, &uinfo);
+       if (err < 0) {
+               pr_err("%s, failed to get iscsi param err=%d.\n",
+                        tdev->name, err);
+               return err;
+       }
+
+       err = cxgbi_ddp_init(cdev, uinfo.llimit, uinfo.ulimit,
+                       uinfo.max_txsz, uinfo.max_rxsz);
+       if (err < 0)
+               return err;
+
+       ddp = cdev->ddp;
+
+       uinfo.tagmask = ddp->idx_mask << PPOD_IDX_SHIFT;
+       cxgbi_ddp_page_size_factor(pgsz_factor);
+       for (i = 0; i < 4; i++)
+               uinfo.pgsz_factor[i] = pgsz_factor[i];
+       uinfo.ulimit = uinfo.llimit + (ddp->nppods << PPOD_SIZE_SHIFT);
+
+       err = tdev->ctl(tdev, ULP_ISCSI_SET_PARAMS, &uinfo);
+       if (err < 0) {
+               pr_warn("%s unable to set iscsi param err=%d, ddp disabled.\n",
+                       tdev->name, err);
+               cxgbi_ddp_cleanup(cdev);
+               return err;
+       }
+       tdev->ulp_iscsi = ddp;
+
+       cdev->csk_ddp_setup_digest = ddp_setup_conn_digest;
+       cdev->csk_ddp_setup_pgidx = ddp_setup_conn_pgidx;
+       cdev->csk_ddp_set = ddp_set_map;
+       cdev->csk_ddp_clear = ddp_clear_map;
+
+       pr_info("tdev 0x%p, nppods %u, bits %u, mask 0x%x,0x%x pkt %u/%u, "
+               "%u/%u.\n",
+               tdev, ddp->nppods, ddp->idx_bits, ddp->idx_mask,
+               ddp->rsvd_tag_mask, ddp->max_txsz, uinfo.max_txsz,
+               ddp->max_rxsz, uinfo.max_rxsz);
+       return 0;
+}
+
+static void cxgb3i_dev_close(struct t3cdev *t3dev)
+{
+       struct cxgbi_device *cdev = cxgbi_device_find_by_lldev(t3dev);
+
+       if (!cdev || cdev->flags & CXGBI_FLAG_ADAPTER_RESET) {
+               pr_info("0x%p close, f 0x%x.\n", cdev, cdev ? cdev->flags : 0);
+               return;
+       }
+
+       cxgbi_device_unregister(cdev);
+}
+
+/**
+ * cxgb3i_dev_open - init a t3 adapter structure and any h/w settings
+ * @t3dev: t3cdev adapter
+ */
+static void cxgb3i_dev_open(struct t3cdev *t3dev)
+{
+       struct cxgbi_device *cdev = cxgbi_device_find_by_lldev(t3dev);
+       struct adapter *adapter = tdev2adap(t3dev);
+       int i, err;
+
+       if (cdev) {
+               pr_info("0x%p, updating.\n", cdev);
+               return;
+       }
+
+       cdev = cxgbi_device_register(0, adapter->params.nports);
+       if (!cdev) {
+               pr_warn("device 0x%p register failed.\n", t3dev);
+               return;
+       }
+
+       cdev->flags = CXGBI_FLAG_DEV_T3 | CXGBI_FLAG_IPV4_SET;
+       cdev->lldev = t3dev;
+       cdev->pdev = adapter->pdev;
+       cdev->ports = adapter->port;
+       cdev->nports = adapter->params.nports;
+       cdev->mtus = adapter->params.mtus;
+       cdev->nmtus = NMTUS;
+       cdev->snd_win = cxgb3i_snd_win;
+       cdev->rcv_win = cxgb3i_rcv_win;
+       cdev->rx_credit_thres = cxgb3i_rx_credit_thres;
+       cdev->skb_tx_rsvd = CXGB3I_TX_HEADER_LEN;
+       cdev->skb_rx_extra = sizeof(struct cpl_iscsi_hdr_norss);
+       cdev->dev_ddp_cleanup = t3_ddp_cleanup;
+       cdev->itp = &cxgb3i_iscsi_transport;
+
+       err = cxgb3i_ddp_init(cdev);
+       if (err) {
+               pr_info("0x%p ddp init failed\n", cdev);
+               goto err_out;
+       }
+
+       err = cxgb3i_ofld_init(cdev);
+       if (err) {
+               pr_info("0x%p offload init failed\n", cdev);
+               goto err_out;
+       }
+
+       err = cxgbi_hbas_add(cdev, CXGB3I_MAX_LUN, CXGBI_MAX_CONN,
+                               &cxgb3i_host_template, cxgb3i_stt);
+       if (err)
+               goto err_out;
+
+       for (i = 0; i < cdev->nports; i++)
+               cdev->hbas[i]->ipv4addr =
+                       cxgb3i_get_private_ipv4addr(cdev->ports[i]);
+
+       pr_info("cdev 0x%p, f 0x%x, t3dev 0x%p open, err %d.\n",
+               cdev, cdev ? cdev->flags : 0, t3dev, err);
+       return;
+
+err_out:
+       cxgbi_device_unregister(cdev);
+}
+
+static void cxgb3i_dev_event_handler(struct t3cdev *t3dev, u32 event, u32 port)
+{
+       struct cxgbi_device *cdev = cxgbi_device_find_by_lldev(t3dev);
+
+       log_debug(1 << CXGBI_DBG_TOE,
+               "0x%p, cdev 0x%p, event 0x%x, port 0x%x.\n",
+               t3dev, cdev, event, port);
+       if (!cdev)
+               return;
+
+       switch (event) {
+       case OFFLOAD_STATUS_DOWN:
+               cdev->flags |= CXGBI_FLAG_ADAPTER_RESET;
+               break;
+       case OFFLOAD_STATUS_UP:
+               cdev->flags &= ~CXGBI_FLAG_ADAPTER_RESET;
+               break;
+       }
+}
+
+/**
+ * cxgb3i_init_module - module init entry point
+ *
+ * initialize any driver wide global data structures and register itself
+ *     with the cxgb3 module
+ */
+static int __init cxgb3i_init_module(void)
+{
+       int rc;
+
+       printk(KERN_INFO "%s", version);
+
+       rc = cxgbi_iscsi_init(&cxgb3i_iscsi_transport, &cxgb3i_stt);
+       if (rc < 0)
+               return rc;
+
+       cxgb3_register_client(&t3_client);
+       return 0;
+}
+
+/**
+ * cxgb3i_exit_module - module cleanup/exit entry point
+ *
+ * go through the driver hba list and for each hba, release any resource held.
+ *     and unregisters iscsi transport and the cxgb3 module
+ */
+static void __exit cxgb3i_exit_module(void)
+{
+       cxgb3_unregister_client(&t3_client);
+       cxgbi_device_unregister_all(CXGBI_FLAG_DEV_T3);
+       cxgbi_iscsi_cleanup(&cxgb3i_iscsi_transport, &cxgb3i_stt);
+}
+
+module_init(cxgb3i_init_module);
+module_exit(cxgb3i_exit_module);