[SCSI] cxgb4i v5: iscsi driver
authorkxie@chelsio.com <kxie@chelsio.com>
Tue, 17 Aug 2010 03:55:53 +0000 (20:55 -0700)
committerJames Bottomley <James.Bottomley@suse.de>
Sun, 5 Sep 2010 17:29:07 +0000 (14:29 -0300)
Added cxgb4i iSCSI driver.

This patch implements the cxgb4i iscsi connection acceleration for the
open-iscsi initiator.

The cxgb4i driver offers the iscsi PDU based offload:
- digest insertion and verification
- payload direct-placement into host memory buffer.

Signed-off-by: Karen Xie <kxie@chelsio.com>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
drivers/scsi/Kconfig
drivers/scsi/Makefile
drivers/scsi/cxgbi/Kconfig [new file with mode: 0644]
drivers/scsi/cxgbi/Makefile [new file with mode: 0644]
drivers/scsi/cxgbi/cxgb4i/Kbuild [new file with mode: 0644]
drivers/scsi/cxgbi/cxgb4i/Kconfig [new file with mode: 0644]
drivers/scsi/cxgbi/cxgb4i/cxgb4i.c [new file with mode: 0644]
drivers/scsi/cxgbi/cxgb4i/cxgb4i.h [new file with mode: 0644]

index bbf91aec64f555e532919f680a4702c99fcfd9b2..bee3aef230dcda684f99babf3d0c29682c42c34d 100644 (file)
@@ -379,6 +379,7 @@ config ISCSI_BOOT_SYSFS
          say Y. Otherwise, say N.
 
 source "drivers/scsi/cxgb3i/Kconfig"
+source "drivers/scsi/cxgbi/Kconfig"
 source "drivers/scsi/bnx2i/Kconfig"
 source "drivers/scsi/be2iscsi/Kconfig"
 
index 2703c6ec5e369a3464e7e2ac3100ff50acf1729f..cb31f8cf09d442a11065cb1974874d5582888235 100644 (file)
@@ -134,6 +134,7 @@ obj-$(CONFIG_SCSI_STEX)             += stex.o
 obj-$(CONFIG_SCSI_MVSAS)       += mvsas/
 obj-$(CONFIG_PS3_ROM)          += ps3rom.o
 obj-$(CONFIG_SCSI_CXGB3_ISCSI) += libiscsi.o libiscsi_tcp.o cxgb3i/
+obj-$(CONFIG_SCSI_CXGB4_ISCSI) += libiscsi.o libiscsi_tcp.o cxgbi/
 obj-$(CONFIG_SCSI_BNX2_ISCSI)  += libiscsi.o bnx2i/
 obj-$(CONFIG_BE2ISCSI)         += libiscsi.o be2iscsi/
 obj-$(CONFIG_SCSI_PMCRAID)     += pmcraid.o
diff --git a/drivers/scsi/cxgbi/Kconfig b/drivers/scsi/cxgbi/Kconfig
new file mode 100644 (file)
index 0000000..a470e38
--- /dev/null
@@ -0,0 +1 @@
+source "drivers/scsi/cxgbi/cxgb4i/Kconfig"
diff --git a/drivers/scsi/cxgbi/Makefile b/drivers/scsi/cxgbi/Makefile
new file mode 100644 (file)
index 0000000..9e8f604
--- /dev/null
@@ -0,0 +1 @@
+obj-$(CONFIG_SCSI_CXGB4_ISCSI) += libcxgbi.o cxgb4i/
diff --git a/drivers/scsi/cxgbi/cxgb4i/Kbuild b/drivers/scsi/cxgbi/cxgb4i/Kbuild
new file mode 100644 (file)
index 0000000..b9f4af7
--- /dev/null
@@ -0,0 +1,3 @@
+EXTRA_CFLAGS += -I$(srctree)/drivers/net/cxgb4
+
+obj-$(CONFIG_SCSI_CXGB4_ISCSI) += cxgb4i.o
diff --git a/drivers/scsi/cxgbi/cxgb4i/Kconfig b/drivers/scsi/cxgbi/cxgb4i/Kconfig
new file mode 100644 (file)
index 0000000..bb94b39
--- /dev/null
@@ -0,0 +1,7 @@
+config SCSI_CXGB4_ISCSI
+       tristate "Chelsio T4 iSCSI support"
+       depends on CHELSIO_T4_DEPENDS
+       select CHELSIO_T4
+       select SCSI_ISCSI_ATTRS
+       ---help---
+         This driver supports iSCSI offload for the Chelsio T4 devices.
diff --git a/drivers/scsi/cxgbi/cxgb4i/cxgb4i.c b/drivers/scsi/cxgbi/cxgb4i/cxgb4i.c
new file mode 100644 (file)
index 0000000..b375a68
--- /dev/null
@@ -0,0 +1,1600 @@
+/*
+ * cxgb4i.c: Chelsio T4 iSCSI driver.
+ *
+ * Copyright (c) 2010 Chelsio Communications, Inc.
+ *
+ * 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: Karen Xie (kxie@chelsio.com)
+ *             Rakesh Ranjan (rranjan@chelsio.com)
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ":%s: " fmt, __func__
+
+#include <linux/version.h>
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <scsi/scsi_host.h>
+#include <net/tcp.h>
+#include <net/dst.h>
+#include <linux/netdevice.h>
+
+#include "t4_msg.h"
+#include "cxgb4.h"
+#include "cxgb4_uld.h"
+#include "t4fw_api.h"
+#include "l2t.h"
+#include "cxgb4i.h"
+
+static unsigned int dbg_level;
+
+#include "../libcxgbi.h"
+
+#define        DRV_MODULE_NAME         "cxgb4i"
+#define DRV_MODULE_DESC                "Chelsio T4 iSCSI Driver"
+#define        DRV_MODULE_VERSION      "0.9.0"
+#define        DRV_MODULE_RELDATE      "May 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 cxgb4i_rcv_win = 256 * 1024;
+module_param(cxgb4i_rcv_win, int, 0644);
+MODULE_PARM_DESC(cxgb4i_rcv_win, "TCP reveive window in bytes");
+
+static int cxgb4i_snd_win = 128 * 1024;
+module_param(cxgb4i_snd_win, int, 0644);
+MODULE_PARM_DESC(cxgb4i_snd_win, "TCP send window in bytes");
+
+static int cxgb4i_rx_credit_thres = 10 * 1024;
+module_param(cxgb4i_rx_credit_thres, int, 0644);
+MODULE_PARM_DESC(cxgb4i_rx_credit_thres,
+               "RX credits return threshold in bytes (default=10KB)");
+
+static unsigned int cxgb4i_max_connect = (8 * 1024);
+module_param(cxgb4i_max_connect, uint, 0644);
+MODULE_PARM_DESC(cxgb4i_max_connect, "Maximum number of connections");
+
+static unsigned short cxgb4i_sport_base = 20000;
+module_param(cxgb4i_sport_base, ushort, 0644);
+MODULE_PARM_DESC(cxgb4i_sport_base, "Starting port number (default 20000)");
+
+typedef void (*cxgb4i_cplhandler_func)(struct cxgbi_device *, struct sk_buff *);
+
+static void *t4_uld_add(const struct cxgb4_lld_info *);
+static int t4_uld_rx_handler(void *, const __be64 *, const struct pkt_gl *);
+static int t4_uld_state_change(void *, enum cxgb4_state state);
+
+static const struct cxgb4_uld_info cxgb4i_uld_info = {
+       .name = DRV_MODULE_NAME,
+       .add = t4_uld_add,
+       .rx_handler = t4_uld_rx_handler,
+       .state_change = t4_uld_state_change,
+};
+
+static struct scsi_host_template cxgb4i_host_template = {
+       .module         = THIS_MODULE,
+       .name           = DRV_MODULE_NAME,
+       .proc_name      = DRV_MODULE_NAME,
+       .can_queue      = CXGB4I_SCSI_HOST_QDEPTH,
+       .queuecommand   = iscsi_queuecommand,
+       .change_queue_depth = iscsi_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,
+};
+
+static struct iscsi_transport cxgb4i_iscsi_transport = {
+       .owner          = THIS_MODULE,
+       .name           = DRV_MODULE_NAME,
+       .caps           = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST |
+                               CAP_DATADGST | CAP_DIGEST_OFFLOAD |
+                               CAP_PADDING_OFFLOAD,
+       .param_mask     = ISCSI_MAX_RECV_DLENGTH | ISCSI_MAX_XMIT_DLENGTH |
+                               ISCSI_HDRDGST_EN | ISCSI_DATADGST_EN |
+                               ISCSI_INITIAL_R2T_EN | ISCSI_MAX_R2T |
+                               ISCSI_IMM_DATA_EN | ISCSI_FIRST_BURST |
+                               ISCSI_MAX_BURST | ISCSI_PDU_INORDER_EN |
+                               ISCSI_DATASEQ_INORDER_EN | ISCSI_ERL |
+                               ISCSI_CONN_PORT | ISCSI_CONN_ADDRESS |
+                               ISCSI_EXP_STATSN | ISCSI_PERSISTENT_PORT |
+                               ISCSI_PERSISTENT_ADDRESS |
+                               ISCSI_TARGET_NAME | ISCSI_TPGT |
+                               ISCSI_USERNAME | ISCSI_PASSWORD |
+                               ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
+                               ISCSI_FAST_ABORT | ISCSI_ABORT_TMO |
+                               ISCSI_LU_RESET_TMO | ISCSI_TGT_RESET_TMO |
+                               ISCSI_PING_TMO | ISCSI_RECV_TMO |
+                               ISCSI_IFACE_NAME | ISCSI_INITIATOR_NAME,
+       .host_param_mask        = ISCSI_HOST_HWADDRESS | ISCSI_HOST_IPADDRESS |
+                               ISCSI_HOST_INITIATOR_NAME |
+                               ISCSI_HOST_NETDEV_NAME,
+       .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 = cxgbi_get_conn_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 */
+       .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 *cxgb4i_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.
+ */
+#define DIV_ROUND_UP(n, d)     (((n) + (d) - 1) / (d))
+#define RCV_BUFSIZ_MASK                0x3FFU
+#define MAX_IMM_TX_PKT_LEN     128
+
+static inline void set_queue(struct sk_buff *skb, unsigned int queue,
+                               const struct cxgbi_sock *csk)
+{
+       skb->queue_mapping = queue;
+}
+
+static int push_tx_frames(struct cxgbi_sock *, int);
+
+/*
+ * is_ofld_imm - check whether a packet can be sent as immediate data
+ * @skb: the packet
+ *
+ * Returns true if a packet can be sent as an offload WR with immediate
+ * data.  We currently use the same limit as for Ethernet packets.
+ */
+static inline int is_ofld_imm(const struct sk_buff *skb)
+{
+       return skb->len <= (MAX_IMM_TX_PKT_LEN -
+                       sizeof(struct fw_ofld_tx_data_wr));
+}
+
+static void send_act_open_req(struct cxgbi_sock *csk, struct sk_buff *skb,
+                               struct l2t_entry *e)
+{
+       struct cpl_act_open_req *req;
+       int wscale = cxgbi_sock_compute_wscale(csk->mss_idx);
+       unsigned long long opt0;
+       unsigned int opt2;
+       unsigned int qid_atid = ((unsigned int)csk->atid) |
+                                (((unsigned int)csk->rss_qid) << 14);
+
+       opt0 = KEEP_ALIVE(1) |
+               WND_SCALE(wscale) |
+               MSS_IDX(csk->mss_idx) |
+               L2T_IDX(((struct l2t_entry *)csk->l2t)->idx) |
+               TX_CHAN(csk->tx_chan) |
+               SMAC_SEL(csk->smac_idx) |
+               ULP_MODE(ULP_MODE_ISCSI) |
+               RCV_BUFSIZ(cxgb4i_rcv_win >> 10);
+       opt2 = RX_CHANNEL(0) |
+               RSS_QUEUE_VALID |
+               (1 << 20) | (1 << 22) |
+               RSS_QUEUE(csk->rss_qid);
+
+       set_wr_txq(skb, CPL_PRIORITY_SETUP, csk->port_id);
+       req = (struct cpl_act_open_req *)skb->head;
+
+       INIT_TP_WR(req, 0);
+       OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
+                                       qid_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->opt0 = cpu_to_be64(opt0);
+       req->params = 0;
+       req->opt2 = cpu_to_be32(opt2);
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p, %pI4:%u-%pI4:%u, atid %d, qid %u.\n",
+               csk, &req->local_ip, ntohs(req->local_port),
+               &req->peer_ip, ntohs(req->peer_port),
+               csk->atid, csk->rss_qid);
+
+       cxgb4_l2t_send(csk->cdev->ports[csk->port_id], skb, csk->l2t);
+}
+
+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, tid %u.\n",
+               csk, csk->state, csk->flags, csk->tid);
+       csk->cpl_close = NULL;
+       set_wr_txq(skb, CPL_PRIORITY_DATA, csk->port_id);
+       INIT_TP_WR(req, tid);
+       OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, tid));
+       req->rsvd = 0;
+
+       cxgbi_sock_skb_entail(csk, skb);
+       if (csk->state >= CTP_ESTABLISHED)
+               push_tx_frames(csk, 1);
+}
+
+static void abort_arp_failure(void *handle, struct sk_buff *skb)
+{
+       struct cxgbi_sock *csk = (struct cxgbi_sock *)handle;
+       struct cpl_abort_req *req;
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p,%u,0x%lx, tid %u, abort.\n",
+               csk, csk->state, csk->flags, csk->tid);
+       req = (struct cpl_abort_req *)skb->data;
+       req->cmd = CPL_ABORT_NO_RST;
+       cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
+}
+
+static void send_abort_req(struct cxgbi_sock *csk)
+{
+       struct cpl_abort_req *req;
+       struct sk_buff *skb = csk->cpl_abort_req;
+
+       if (unlikely(csk->state == CTP_ABORTING) || !skb || !csk->cdev)
+               return;
+       cxgbi_sock_set_state(csk, CTP_ABORTING);
+       cxgbi_sock_set_flag(csk, CTPF_ABORT_RPL_PENDING);
+       cxgbi_sock_purge_write_queue(csk);
+
+       csk->cpl_abort_req = NULL;
+       req = (struct cpl_abort_req *)skb->head;
+       set_queue(skb, CPL_PRIORITY_DATA, csk);
+       req->cmd = CPL_ABORT_SEND_RST;
+       t4_set_arp_err_handler(skb, csk, abort_arp_failure);
+       INIT_TP_WR(req, csk->tid);
+       OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, csk->tid));
+       req->rsvd0 = htonl(csk->snd_nxt);
+       req->rsvd1 = !cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT);
+
+       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);
+
+       cxgb4_l2t_send(csk->cdev->ports[csk->port_id], skb, csk->l2t);
+}
+
+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;
+       set_queue(skb, CPL_PRIORITY_DATA, csk);
+       INIT_TP_WR(rpl, csk->tid);
+       OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, csk->tid));
+       rpl->cmd = rst_status;
+       cxgb4_ofld_send(csk->cdev->ports[csk->port_id], 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;
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
+               "csk 0x%p,%u,0x%lx,%u, credit %u.\n",
+               csk, csk->state, csk->flags, csk->tid, credits);
+
+       skb = alloc_cpl(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;
+
+       set_wr_txq(skb, CPL_PRIORITY_ACK, csk->port_id);
+       INIT_TP_WR(req, csk->tid);
+       OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
+                                     csk->tid));
+       req->credit_dack = cpu_to_be32(RX_CREDITS(credits) | RX_FORCE_ACK(1));
+       cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
+       return credits;
+}
+
+/*
+ * sgl_len - calculates the size of an SGL of the given capacity
+ * @n: the number of SGL entries
+ * Calculates the number of flits needed for a scatter/gather list that
+ * can hold the given number of entries.
+ */
+static inline unsigned int sgl_len(unsigned int n)
+{
+       n--;
+       return (3 * n) / 2 + (n & 1) + 2;
+}
+
+/*
+ * calc_tx_flits_ofld - calculate # of flits for an offload packet
+ * @skb: the packet
+ *
+ * Returns the number of flits needed for the given offload packet.
+ * These packets are already fully constructed and no additional headers
+ * will be added.
+ */
+static inline unsigned int calc_tx_flits_ofld(const struct sk_buff *skb)
+{
+       unsigned int flits, cnt;
+
+       if (is_ofld_imm(skb))
+               return DIV_ROUND_UP(skb->len, 8);
+       flits = skb_transport_offset(skb) / 8;
+       cnt = skb_shinfo(skb)->nr_frags;
+       if (skb->tail != skb->transport_header)
+               cnt++;
+       return flits + sgl_len(cnt);
+}
+
+static inline void send_tx_flowc_wr(struct cxgbi_sock *csk)
+{
+       struct sk_buff *skb;
+       struct fw_flowc_wr *flowc;
+       int flowclen, i;
+
+       flowclen = 80;
+       skb = alloc_cpl(flowclen, 0, GFP_ATOMIC);
+       flowc = (struct fw_flowc_wr *)skb->head;
+       flowc->op_to_nparams =
+               htonl(FW_WR_OP(FW_FLOWC_WR) | FW_FLOWC_WR_NPARAMS(8));
+       flowc->flowid_len16 =
+               htonl(FW_WR_LEN16(DIV_ROUND_UP(72, 16)) |
+                               FW_WR_FLOWID(csk->tid));
+       flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
+       flowc->mnemval[0].val = htonl(0);
+       flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
+       flowc->mnemval[1].val = htonl(csk->tx_chan);
+       flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
+       flowc->mnemval[2].val = htonl(csk->tx_chan);
+       flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
+       flowc->mnemval[3].val = htonl(csk->rss_qid);
+       flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
+       flowc->mnemval[4].val = htonl(csk->snd_nxt);
+       flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
+       flowc->mnemval[5].val = htonl(csk->rcv_nxt);
+       flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
+       flowc->mnemval[6].val = htonl(cxgb4i_snd_win);
+       flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
+       flowc->mnemval[7].val = htonl(csk->advmss);
+       flowc->mnemval[8].mnemonic = 0;
+       flowc->mnemval[8].val = 0;
+       for (i = 0; i < 9; i++) {
+               flowc->mnemval[i].r4[0] = 0;
+               flowc->mnemval[i].r4[1] = 0;
+               flowc->mnemval[i].r4[2] = 0;
+       }
+       set_queue(skb, CPL_PRIORITY_DATA, csk);
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p, tid 0x%x, %u,%u,%u,%u,%u,%u,%u.\n",
+               csk, csk->tid, 0, csk->tx_chan, csk->rss_qid,
+               csk->snd_nxt, csk->rcv_nxt, cxgb4i_snd_win,
+               csk->advmss);
+
+       cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
+}
+
+static inline void make_tx_data_wr(struct cxgbi_sock *csk, struct sk_buff *skb,
+                                  int dlen, int len, u32 credits, int compl)
+{
+       struct fw_ofld_tx_data_wr *req;
+       unsigned int submode = cxgbi_skcb_ulp_mode(skb) & 3;
+       unsigned int wr_ulp_mode = 0;
+
+       req = (struct fw_ofld_tx_data_wr *)__skb_push(skb, sizeof(*req));
+
+       if (is_ofld_imm(skb)) {
+               req->op_to_immdlen = htonl(FW_WR_OP(FW_OFLD_TX_DATA_WR) |
+                                       FW_WR_COMPL(1) |
+                                       FW_WR_IMMDLEN(dlen));
+               req->flowid_len16 = htonl(FW_WR_FLOWID(csk->tid) |
+                                               FW_WR_LEN16(credits));
+       } else {
+               req->op_to_immdlen =
+                       cpu_to_be32(FW_WR_OP(FW_OFLD_TX_DATA_WR) |
+                                       FW_WR_COMPL(1) |
+                                       FW_WR_IMMDLEN(0));
+               req->flowid_len16 =
+                       cpu_to_be32(FW_WR_FLOWID(csk->tid) |
+                                       FW_WR_LEN16(credits));
+       }
+       if (submode)
+               wr_ulp_mode = FW_OFLD_TX_DATA_WR_ULPMODE(ULP2_MODE_ISCSI) |
+                               FW_OFLD_TX_DATA_WR_ULPSUBMODE(submode);
+       req->tunnel_to_proxy = htonl(wr_ulp_mode) |
+                FW_OFLD_TX_DATA_WR_SHOVE(skb_peek(&csk->write_queue) ? 0 : 1);
+       req->plen = htonl(len);
+       if (!cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT))
+               cxgbi_sock_set_flag(csk, CTPF_TX_DATA_SENT);
+}
+
+static void arp_failure_skb_discard(void *handle, 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_SOCK |
+                        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 dlen = skb->len;
+               int len = skb->len;
+               unsigned int credits_needed;
+
+               skb_reset_transport_header(skb);
+               if (is_ofld_imm(skb))
+                       credits_needed = DIV_ROUND_UP(dlen +
+                                       sizeof(struct fw_ofld_tx_data_wr), 16);
+               else
+                       credits_needed = DIV_ROUND_UP(8*calc_tx_flits_ofld(skb)
+                                       + sizeof(struct fw_ofld_tx_data_wr),
+                                       16);
+
+               if (csk->wr_cred < credits_needed) {
+                       log_debug(1 << CXGBI_DBG_PDU_TX,
+                               "csk 0x%p, skb %u/%u, wr %d < %u.\n",
+                               csk, skb->len, skb->data_len,
+                               credits_needed, csk->wr_cred);
+                       break;
+               }
+               __skb_unlink(skb, &csk->write_queue);
+               set_queue(skb, CPL_PRIORITY_DATA, csk);
+               skb->csum = credits_needed;
+               csk->wr_cred -= credits_needed;
+               csk->wr_una_cred += credits_needed;
+               cxgbi_sock_enqueue_wr(csk, skb);
+
+               log_debug(1 << CXGBI_DBG_PDU_TX,
+                       "csk 0x%p, skb %u/%u, wr %d, left %u, unack %u.\n",
+                       csk, skb->len, skb->data_len, credits_needed,
+                       csk->wr_cred, csk->wr_una_cred);
+
+               if (likely(cxgbi_skcb_test_flag(skb, SKCBF_TX_NEED_HDR))) {
+                       if (!cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT)) {
+                               send_tx_flowc_wr(csk);
+                               skb->csum += 5;
+                               csk->wr_cred -= 5;
+                               csk->wr_una_cred += 5;
+                       }
+                       len += cxgbi_ulp_extra_len(cxgbi_skcb_ulp_mode(skb));
+                       make_tx_data_wr(csk, skb, dlen, len, credits_needed,
+                                       req_completion);
+                       csk->snd_nxt += len;
+                       cxgbi_skcb_clear_flag(skb, SKCBF_TX_NEED_HDR);
+               }
+               total_size += skb->truesize;
+               t4_set_arp_err_handler(skb, csk, arp_failure_skb_discard);
+
+               log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_TX,
+                       "csk 0x%p,%u,0x%lx,%u, skb 0x%p, %u.\n",
+                       csk, csk->state, csk->flags, csk->tid, skb, len);
+
+               cxgb4_l2t_send(csk->cdev->ports[csk->port_id], skb, csk->l2t);
+       }
+       return total_size;
+}
+
+static inline void free_atid(struct cxgbi_sock *csk)
+{
+       struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(csk->cdev);
+
+       if (cxgbi_sock_flag(csk, CTPF_HAS_ATID)) {
+               cxgb4_free_atid(lldi->tids, csk->atid);
+               cxgbi_sock_clear_flag(csk, CTPF_HAS_ATID);
+               cxgbi_sock_put(csk);
+       }
+}
+
+static void do_act_establish(struct cxgbi_device *cdev, struct sk_buff *skb)
+{
+       struct cxgbi_sock *csk;
+       struct cpl_act_establish *req = (struct cpl_act_establish *)skb->data;
+       unsigned short tcp_opt = ntohs(req->tcp_opt);
+       unsigned int tid = GET_TID(req);
+       unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
+       struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
+       struct tid_info *t = lldi->tids;
+       u32 rcv_isn = be32_to_cpu(req->rcv_isn);
+
+       csk = lookup_atid(t, atid);
+       if (unlikely(!csk)) {
+               pr_err("NO conn. for atid %u, cdev 0x%p.\n", atid, cdev);
+               goto rel_skb;
+       }
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p,%u,0x%lx, tid %u, atid %u, rseq %u.\n",
+               csk, csk->state, csk->flags, tid, atid, rcv_isn);
+
+       cxgbi_sock_get(csk);
+       csk->tid = tid;
+       cxgb4_insert_tid(lldi->tids, csk, tid);
+       cxgbi_sock_set_flag(csk, CTPF_HAS_TID);
+
+       free_atid(csk);
+
+       spin_lock_bh(&csk->lock);
+       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);
+
+       if (csk->retry_timer.function) {
+               del_timer(&csk->retry_timer);
+               csk->retry_timer.function = NULL;
+       }
+
+       csk->copied_seq = csk->rcv_wup = csk->rcv_nxt = rcv_isn;
+       /*
+        * Causes the first RX_DATA_ACK to supply any Rx credits we couldn't
+        * pass through opt0.
+        */
+       if (cxgb4i_rcv_win > (RCV_BUFSIZ_MASK << 10))
+               csk->rcv_wup -= cxgb4i_rcv_win - (RCV_BUFSIZ_MASK << 10);
+
+       csk->advmss = lldi->mtus[GET_TCPOPT_MSS(tcp_opt)] - 40;
+       if (GET_TCPOPT_TSTAMP(tcp_opt))
+               csk->advmss -= 12;
+       if (csk->advmss < 128)
+               csk->advmss = 128;
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p, mss_idx %u, advmss %u.\n",
+                       csk, GET_TCPOPT_MSS(tcp_opt), csk->advmss);
+
+       cxgbi_sock_established(csk, ntohl(req->snd_isn), ntohs(req->tcp_opt));
+
+       if (unlikely(cxgbi_sock_flag(csk, CTPF_ACTIVE_CLOSE_NEEDED)))
+               send_abort_req(csk);
+       else {
+               if (skb_queue_len(&csk->write_queue))
+                       push_tx_frames(csk, 0);
+               cxgbi_conn_tx_open(csk);
+       }
+       spin_unlock_bh(&csk->lock);
+
+rel_skb:
+       __kfree_skb(skb);
+}
+
+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 csk_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_cpl(sizeof(struct cpl_act_open_req), 0, GFP_ATOMIC);
+       if (!skb)
+               cxgbi_sock_fail_act_open(csk, -ENOMEM);
+       else {
+               skb->sk = (struct sock *)csk;
+               t4_set_arp_err_handler(skb, csk,
+                                       cxgbi_sock_act_open_req_arp_failure);
+               send_act_open_req(csk, skb, csk->l2t);
+       }
+       spin_unlock_bh(&csk->lock);
+       cxgbi_sock_put(csk);
+}
+
+static void do_act_open_rpl(struct cxgbi_device *cdev, struct sk_buff *skb)
+{
+       struct cxgbi_sock *csk;
+       struct cpl_act_open_rpl *rpl = (struct cpl_act_open_rpl *)skb->data;
+       unsigned int tid = GET_TID(rpl);
+       unsigned int atid =
+               GET_TID_TID(GET_AOPEN_ATID(be32_to_cpu(rpl->atid_status)));
+       unsigned int status = GET_AOPEN_STATUS(be32_to_cpu(rpl->atid_status));
+       struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
+       struct tid_info *t = lldi->tids;
+
+       csk = lookup_atid(t, atid);
+       if (unlikely(!csk)) {
+               pr_err("NO matching conn. atid %u, tid %u.\n", atid, tid);
+               goto rel_skb;
+       }
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p,%u,0x%lx, status %u, atid %u, tid %u.\n",
+                       csk, csk->state, csk->flags, status, atid, tid);
+
+       if (status && status != CPL_ERR_TCAM_FULL &&
+           status != CPL_ERR_CONN_EXIST &&
+           status != CPL_ERR_ARP_MISS)
+               cxgb4_remove_tid(lldi->tids, csk->port_id, GET_TID(rpl));
+
+       cxgbi_sock_get(csk);
+       spin_lock_bh(&csk->lock);
+
+       if (status == CPL_ERR_CONN_EXIST &&
+           csk->retry_timer.function != csk_act_open_retry_timer) {
+               csk->retry_timer.function = csk_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(status));
+
+       spin_unlock_bh(&csk->lock);
+       cxgbi_sock_put(csk);
+rel_skb:
+       __kfree_skb(skb);
+}
+
+static void do_peer_close(struct cxgbi_device *cdev, struct sk_buff *skb)
+{
+       struct cxgbi_sock *csk;
+       struct cpl_peer_close *req = (struct cpl_peer_close *)skb->data;
+       unsigned int tid = GET_TID(req);
+       struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
+       struct tid_info *t = lldi->tids;
+
+       csk = lookup_tid(t, tid);
+       if (unlikely(!csk)) {
+               pr_err("can't find connection for tid %u.\n", tid);
+               goto rel_skb;
+       }
+       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);
+rel_skb:
+       __kfree_skb(skb);
+}
+
+static void do_close_con_rpl(struct cxgbi_device *cdev, struct sk_buff *skb)
+{
+       struct cxgbi_sock *csk;
+       struct cpl_close_con_rpl *rpl = (struct cpl_close_con_rpl *)skb->data;
+       unsigned int tid = GET_TID(rpl);
+       struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
+       struct tid_info *t = lldi->tids;
+
+       csk = lookup_tid(t, tid);
+       if (unlikely(!csk)) {
+               pr_err("can't find connection for tid %u.\n", tid);
+               goto rel_skb;
+       }
+       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_close_conn_rpl(csk, ntohl(rpl->snd_nxt));
+rel_skb:
+       __kfree_skb(skb);
+}
+
+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 void do_abort_req_rss(struct cxgbi_device *cdev, struct sk_buff *skb)
+{
+       struct cxgbi_sock *csk;
+       struct cpl_abort_req_rss *req = (struct cpl_abort_req_rss *)skb->data;
+       unsigned int tid = GET_TID(req);
+       struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
+       struct tid_info *t = lldi->tids;
+       int rst_status = CPL_ABORT_NO_RST;
+
+       csk = lookup_tid(t, tid);
+       if (unlikely(!csk)) {
+               pr_err("can't find connection for tid %u.\n", tid);
+               goto rel_skb;
+       }
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p,%u,0x%lx, tid %u, status 0x%x.\n",
+               csk, csk->state, csk->flags, csk->tid, req->status);
+
+       if (req->status == CPL_ERR_RTX_NEG_ADVICE ||
+           req->status == CPL_ERR_PERSIST_NEG_ADVICE)
+               goto rel_skb;
+
+       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 done;
+       }
+
+       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);
+       }
+done:
+       spin_unlock_bh(&csk->lock);
+       cxgbi_sock_put(csk);
+rel_skb:
+       __kfree_skb(skb);
+}
+
+static void do_abort_rpl_rss(struct cxgbi_device *cdev, struct sk_buff *skb)
+{
+       struct cxgbi_sock *csk;
+       struct cpl_abort_rpl_rss *rpl = (struct cpl_abort_rpl_rss *)skb->data;
+       unsigned int tid = GET_TID(rpl);
+       struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
+       struct tid_info *t = lldi->tids;
+
+       csk = lookup_tid(t, tid);
+       if (!csk)
+               goto rel_skb;
+
+       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);
+
+       if (rpl->status == CPL_ERR_ABORT_FAILED)
+               goto rel_skb;
+
+       cxgbi_sock_rcv_abort_rpl(csk);
+rel_skb:
+       __kfree_skb(skb);
+}
+
+static void do_rx_iscsi_hdr(struct cxgbi_device *cdev, struct sk_buff *skb)
+{
+       struct cxgbi_sock *csk;
+       struct cpl_iscsi_hdr *cpl = (struct cpl_iscsi_hdr *)skb->data;
+       unsigned short pdu_len_ddp = be16_to_cpu(cpl->pdu_len_ddp);
+       unsigned int tid = GET_TID(cpl);
+       struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
+       struct tid_info *t = lldi->tids;
+       struct sk_buff *lskb;
+
+       csk = lookup_tid(t, tid);
+       if (unlikely(!csk)) {
+               pr_err("can't find conn. for tid %u.\n", tid);
+               goto rel_skb;
+       }
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
+               "csk 0x%p,%u,0x%lx, tid %u, skb 0x%p,%u, 0x%x.\n",
+               csk, csk->state, csk->flags, csk->tid, skb, skb->len,
+               pdu_len_ddp);
+
+       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(cpl->seq);
+       skb_reset_transport_header(skb);
+       __skb_pull(skb, sizeof(*cpl));
+       __pskb_trim(skb, ntohs(cpl->len));
+
+       if (!csk->skb_ulp_lhdr) {
+               unsigned char *bhs;
+               unsigned int hlen, dlen;
+
+               log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
+                       "csk 0x%p,%u,0x%lx, tid %u, skb 0x%p header.\n",
+                       csk, csk->state, csk->flags, csk->tid, skb);
+               csk->skb_ulp_lhdr = skb;
+               lskb = csk->skb_ulp_lhdr;
+               cxgbi_skcb_set_flag(lskb, SKCBF_RX_HDR);
+
+               if (cxgbi_skcb_tcp_seq(lskb) != csk->rcv_nxt) {
+                       pr_info("tid %u, CPL_ISCSI_HDR, bad seq, 0x%x/0x%x.\n",
+                               csk->tid, cxgbi_skcb_tcp_seq(lskb),
+                               csk->rcv_nxt);
+                       goto abort_conn;
+               }
+
+               bhs = lskb->data;
+               hlen = ntohs(cpl->len);
+               dlen = ntohl(*(unsigned int *)(bhs + 4)) & 0xFFFFFF;
+
+               if ((hlen + dlen) != ISCSI_PDU_LEN(pdu_len_ddp) - 40) {
+                       pr_info("tid 0x%x, CPL_ISCSI_HDR, pdu len "
+                               "mismatch %u != %u + %u, seq 0x%x.\n",
+                               csk->tid, ISCSI_PDU_LEN(pdu_len_ddp) - 40,
+                               hlen, dlen, cxgbi_skcb_tcp_seq(skb));
+                       goto abort_conn;
+               }
+
+               cxgbi_skcb_rx_pdulen(skb) = (hlen + dlen + 3) & (~0x3);
+               if (dlen)
+                       cxgbi_skcb_rx_pdulen(skb) += csk->dcrc_len;
+               csk->rcv_nxt += cxgbi_skcb_rx_pdulen(skb);
+
+               log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
+                       "csk 0x%p, skb 0x%p, 0x%x,%u+%u,0x%x,0x%x.\n",
+                       csk, skb, *bhs, hlen, dlen,
+                       ntohl(*((unsigned int *)(bhs + 16))),
+                       ntohl(*((unsigned int *)(bhs + 24))));
+
+       } else {
+               lskb = csk->skb_ulp_lhdr;
+               cxgbi_skcb_set_flag(lskb, SKCBF_RX_DATA);
+
+               log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
+                       "csk 0x%p,%u,0x%lx, skb 0x%p data, 0x%p.\n",
+                       csk, csk->state, csk->flags, skb, lskb);
+       }
+
+       __skb_queue_tail(&csk->receive_queue, skb);
+       spin_unlock_bh(&csk->lock);
+       return;
+
+abort_conn:
+       send_abort_req(csk);
+discard:
+       spin_unlock_bh(&csk->lock);
+rel_skb:
+       __kfree_skb(skb);
+}
+
+static void do_rx_data_ddp(struct cxgbi_device *cdev,
+                                 struct sk_buff *skb)
+{
+       struct cxgbi_sock *csk;
+       struct sk_buff *lskb;
+       struct cpl_rx_data_ddp *rpl = (struct cpl_rx_data_ddp *)skb->data;
+       unsigned int tid = GET_TID(rpl);
+       struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
+       struct tid_info *t = lldi->tids;
+       unsigned int status = ntohl(rpl->ddpvld);
+
+       csk = lookup_tid(t, tid);
+       if (unlikely(!csk)) {
+               pr_err("can't find connection for tid %u.\n", tid);
+               goto rel_skb;
+       }
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
+               "csk 0x%p,%u,0x%lx, skb 0x%p,0x%x, lhdr 0x%p.\n",
+               csk, csk->state, csk->flags, skb, status, csk->skb_ulp_lhdr);
+
+       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;
+       }
+
+       if (!csk->skb_ulp_lhdr) {
+               pr_err("tid 0x%x, rcv RX_DATA_DDP w/o pdu bhs.\n", csk->tid);
+               goto abort_conn;
+       }
+
+       lskb = csk->skb_ulp_lhdr;
+       csk->skb_ulp_lhdr = NULL;
+
+       cxgbi_skcb_set_flag(lskb, SKCBF_RX_STATUS);
+       cxgbi_skcb_rx_ddigest(lskb) = ntohl(rpl->ulp_crc);
+
+       if (ntohs(rpl->len) != cxgbi_skcb_rx_pdulen(lskb))
+               pr_info("tid 0x%x, RX_DATA_DDP pdulen %u != %u.\n",
+                       csk->tid, ntohs(rpl->len), cxgbi_skcb_rx_pdulen(lskb));
+
+       if (status & (1 << CPL_RX_DDP_STATUS_HCRC_SHIFT)) {
+               log_debug(1 << CXGBI_DBG_PDU_RX,
+                       "csk 0x%p, lhdr 0x%p, status 0x%x, hcrc bad.\n",
+                       csk, lskb, status);
+               cxgbi_skcb_set_flag(lskb, SKCBF_RX_HCRC_ERR);
+       }
+       if (status & (1 << CPL_RX_DDP_STATUS_DCRC_SHIFT)) {
+               log_debug(1 << CXGBI_DBG_PDU_RX,
+                       "csk 0x%p, lhdr 0x%p, status 0x%x, dcrc bad.\n",
+                       csk, lskb, status);
+               cxgbi_skcb_set_flag(lskb, SKCBF_RX_DCRC_ERR);
+       }
+       if (status & (1 << CPL_RX_DDP_STATUS_PAD_SHIFT)) {
+               log_debug(1 << CXGBI_DBG_PDU_RX,
+                       "csk 0x%p, lhdr 0x%p, status 0x%x, pad bad.\n",
+                       csk, lskb, status);
+               cxgbi_skcb_set_flag(lskb, SKCBF_RX_PAD_ERR);
+       }
+       if ((status & (1 << CPL_RX_DDP_STATUS_DDP_SHIFT)) &&
+               !cxgbi_skcb_test_flag(lskb, SKCBF_RX_DATA)) {
+               log_debug(1 << CXGBI_DBG_PDU_RX,
+                       "csk 0x%p, lhdr 0x%p, 0x%x, data ddp'ed.\n",
+                       csk, lskb, status);
+               cxgbi_skcb_set_flag(lskb, SKCBF_RX_DATA_DDPD);
+       }
+       log_debug(1 << CXGBI_DBG_PDU_RX,
+               "csk 0x%p, lskb 0x%p, f 0x%lx.\n",
+               csk, lskb, cxgbi_skcb_flags(lskb));
+
+       cxgbi_conn_pdu_ready(csk);
+       spin_unlock_bh(&csk->lock);
+       goto rel_skb;
+
+abort_conn:
+       send_abort_req(csk);
+discard:
+       spin_unlock_bh(&csk->lock);
+rel_skb:
+       __kfree_skb(skb);
+}
+
+static void do_fw4_ack(struct cxgbi_device *cdev, struct sk_buff *skb)
+{
+       struct cxgbi_sock *csk;
+       struct cpl_fw4_ack *rpl = (struct cpl_fw4_ack *)skb->data;
+       unsigned int tid = GET_TID(rpl);
+       struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
+       struct tid_info *t = lldi->tids;
+
+       csk = lookup_tid(t, tid);
+       if (unlikely(!csk))
+               pr_err("can't find connection for tid %u.\n", tid);
+       else {
+               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_wr_ack(csk, rpl->credits, ntohl(rpl->snd_una),
+                                       rpl->seq_vld);
+       }
+       __kfree_skb(skb);
+}
+
+static void do_set_tcb_rpl(struct cxgbi_device *cdev, struct sk_buff *skb)
+{
+       struct cpl_set_tcb_rpl *rpl = (struct cpl_set_tcb_rpl *)skb->data;
+       unsigned int tid = GET_TID(rpl);
+       struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
+       struct tid_info *t = lldi->tids;
+       struct cxgbi_sock *csk;
+
+       csk = lookup_tid(t, tid);
+       if (!csk)
+               pr_err("can't find conn. for tid %u.\n", tid);
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p,%u,%lx,%u, status 0x%x.\n",
+               csk, csk->state, csk->flags, csk->tid, rpl->status);
+
+       if (rpl->status != CPL_ERR_NONE)
+               pr_err("csk 0x%p,%u, SET_TCB_RPL status %u.\n",
+                       csk, tid, rpl->status);
+
+       __kfree_skb(skb);
+}
+
+static int alloc_cpls(struct cxgbi_sock *csk)
+{
+       csk->cpl_close = alloc_cpl(sizeof(struct cpl_close_con_req),
+                                       0, GFP_NOIO);
+       if (!csk->cpl_close)
+               return -ENOMEM;
+
+       csk->cpl_abort_req = alloc_cpl(sizeof(struct cpl_abort_req),
+                                       0, GFP_NOIO);
+       if (!csk->cpl_abort_req)
+               goto free_cpls;
+
+       csk->cpl_abort_rpl = alloc_cpl(sizeof(struct cpl_abort_rpl),
+                                       0, GFP_NOIO);
+       if (!csk->cpl_abort_rpl)
+               goto free_cpls;
+       return 0;
+
+free_cpls:
+       cxgbi_sock_free_cpl_skbs(csk);
+       return -ENOMEM;
+}
+
+static inline void l2t_put(struct cxgbi_sock *csk)
+{
+       if (csk->l2t) {
+               cxgb4_l2t_release(csk->l2t);
+               csk->l2t = NULL;
+               cxgbi_sock_put(csk);
+       }
+}
+
+static void release_offload_resources(struct cxgbi_sock *csk)
+{
+       struct cxgb4_lld_info *lldi;
+
+       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_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)) {
+               lldi = cxgbi_cdev_priv(csk->cdev);
+               cxgb4_remove_tid(lldi->tids, 0, csk->tid);
+               cxgbi_sock_clear_flag(csk, CTPF_HAS_TID);
+               cxgbi_sock_put(csk);
+       }
+       csk->dst = NULL;
+       csk->cdev = NULL;
+}
+
+static int init_act_open(struct cxgbi_sock *csk)
+{
+       struct cxgbi_device *cdev = csk->cdev;
+       struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
+       struct net_device *ndev = cdev->ports[csk->port_id];
+       struct port_info *pi = netdev_priv(ndev);
+       struct sk_buff *skb = NULL;
+       unsigned int step;
+
+       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->atid = cxgb4_alloc_atid(lldi->tids, csk);
+       if (csk->atid < 0) {
+               pr_err("%s, NO atid available.\n", ndev->name);
+               return -EINVAL;
+       }
+       cxgbi_sock_set_flag(csk, CTPF_HAS_ATID);
+       cxgbi_sock_get(csk);
+
+       csk->l2t = cxgb4_l2t_get(lldi->l2t, csk->dst->neighbour, ndev, 0);
+       if (!csk->l2t) {
+               pr_err("%s, cannot alloc l2t.\n", ndev->name);
+               goto rel_resource;
+       }
+       cxgbi_sock_get(csk);
+
+       skb = alloc_cpl(sizeof(struct cpl_act_open_req), 0, GFP_NOIO);
+       if (!skb)
+               goto rel_resource;
+       skb->sk = (struct sock *)csk;
+       t4_set_arp_err_handler(skb, csk, cxgbi_sock_act_open_req_arp_failure);
+
+       if (!csk->mtu)
+               csk->mtu = dst_mtu(csk->dst);
+       cxgb4_best_mtu(lldi->mtus, csk->mtu, &csk->mss_idx);
+       csk->tx_chan = cxgb4_port_chan(ndev);
+       /* SMT two entries per row */
+       csk->smac_idx = ((cxgb4_port_viid(ndev) & 0x7F)) << 1;
+       step = lldi->ntxq / lldi->nchan;
+       csk->txq_idx = cxgb4_port_idx(ndev) * step;
+       step = lldi->nrxq / lldi->nchan;
+       csk->rss_qid = lldi->rxq_ids[cxgb4_port_idx(ndev) * step];
+       csk->wr_max_cred = csk->wr_cred = lldi->wr_cred;
+       csk->wr_una_cred = 0;
+       cxgbi_sock_reset_wr_list(csk);
+       csk->err = 0;
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p,p%d,%s, %u,%u,%u, mss %u,%u, smac %u.\n",
+               csk, pi->port_id, ndev->name, csk->tx_chan,
+               csk->txq_idx, csk->rss_qid, csk->mtu, csk->mss_idx,
+               csk->smac_idx);
+
+       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;
+}
+
+cxgb4i_cplhandler_func cxgb4i_cplhandlers[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_rss,
+       [CPL_ABORT_RPL_RSS] = do_abort_rpl_rss,
+       [CPL_CLOSE_CON_RPL] = do_close_con_rpl,
+       [CPL_FW4_ACK] = do_fw4_ack,
+       [CPL_ISCSI_HDR] = do_rx_iscsi_hdr,
+       [CPL_SET_TCB_RPL] = do_set_tcb_rpl,
+       [CPL_RX_DATA_DDP] = do_rx_data_ddp,
+};
+
+int cxgb4i_ofld_init(struct cxgbi_device *cdev)
+{
+       int rc;
+
+       if (cxgb4i_max_connect > CXGB4I_MAX_CONN)
+               cxgb4i_max_connect = CXGB4I_MAX_CONN;
+
+       rc = cxgbi_device_portmap_create(cdev, cxgb4i_sport_base,
+                                       cxgb4i_max_connect);
+       if (rc < 0)
+               return rc;
+
+       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 ulp_mem_io *req,
+                               unsigned int dlen, unsigned int pm_addr)
+{
+       struct ulptx_sgl *sgl;
+       unsigned int wr_len = roundup(sizeof(struct ulp_mem_io) +
+                                       sizeof(struct ulptx_sgl), 16);
+
+       INIT_ULPTX_WR(req, wr_len, 0, 0);
+       req->cmd = htonl(ULPTX_CMD(ULP_TX_MEM_WRITE));
+       req->dlen = htonl(ULP_MEMIO_DATA_LEN(dlen >> 5));
+       req->lock_addr = htonl(ULP_MEMIO_ADDR(pm_addr >> 5));
+       req->len16 = htonl(DIV_ROUND_UP(wr_len - sizeof(req->wr), 16));
+       sgl = (struct ulptx_sgl *)(req + 1);
+       sgl->cmd_nsge = htonl(ULPTX_CMD(ULP_TX_SC_DSGL) | ULPTX_NSGE(1));
+       sgl->len0 = htonl(dlen);
+}
+
+static int ddp_ppod_write_sgl(struct cxgbi_device *cdev, unsigned int port_id,
+                               struct cxgbi_pagepod_hdr *hdr, unsigned int idx,
+                               unsigned int npods,
+                               struct cxgbi_gather_list *gl,
+                               unsigned int gl_pidx)
+{
+       struct cxgbi_ddp_info *ddp = cdev->ddp;
+       unsigned int dlen, pm_addr;
+       struct sk_buff *skb;
+       struct ulp_mem_io *req;
+       struct ulptx_sgl *sgl;
+       struct cxgbi_pagepod *ppod;
+       unsigned int i;
+
+       dlen = PPOD_SIZE * npods;
+       pm_addr = idx * PPOD_SIZE + ddp->llimit;
+
+       skb = alloc_cpl(sizeof(*req) + sizeof(*sgl), dlen, GFP_ATOMIC);
+       if (!skb) {
+               pr_err("cdev 0x%p, idx %u, npods %u, OOM.\n",
+                       cdev, idx, npods);
+               return -ENOMEM;
+       }
+       req = (struct ulp_mem_io *)skb->head;
+       set_queue(skb, CPL_PRIORITY_CONTROL, NULL);
+
+       ulp_mem_io_set_hdr(req, dlen, pm_addr);
+       sgl = (struct ulptx_sgl *)(req + 1);
+       ppod = (struct cxgbi_pagepod *)(sgl + 1);
+       sgl->addr0 = cpu_to_be64(virt_to_phys(ppod));
+
+       for (i = 0; i < npods; i++, ppod++, gl_pidx += PPOD_PAGES_MAX) {
+               if (!hdr && !gl)
+                       cxgbi_ddp_ppod_clear(ppod);
+               else
+                       cxgbi_ddp_ppod_set(ppod, hdr, gl, gl_pidx);
+       }
+
+       cxgb4_ofld_send(cdev->ports[port_id], skb);
+       return 0;
+}
+
+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)
+{
+       unsigned int i, cnt;
+       int err = 0;
+
+       for (i = 0; i < npods; i += cnt, idx += cnt) {
+               cnt = npods - i;
+               if (cnt > ULPMEM_DSGL_MAX_NPPODS)
+                       cnt = ULPMEM_DSGL_MAX_NPPODS;
+               err = ddp_ppod_write_sgl(csk->cdev, csk->port_id, hdr,
+                                       idx, cnt, gl, 4 * i);
+               if (err < 0)
+                       break;
+       }
+       return err;
+}
+
+static void ddp_clear_map(struct cxgbi_hba *chba, unsigned int tag,
+                         unsigned int idx, unsigned int npods)
+{
+       unsigned int i, cnt;
+       int err;
+
+       for (i = 0; i < npods; i += cnt, idx += cnt) {
+               cnt = npods - i;
+               if (cnt > ULPMEM_DSGL_MAX_NPPODS)
+                       cnt = ULPMEM_DSGL_MAX_NPPODS;
+               err = ddp_ppod_write_sgl(chba->cdev, chba->port_id, NULL,
+                                       idx, cnt, NULL, 0);
+               if (err < 0)
+                       break;
+       }
+}
+
+static int ddp_setup_conn_pgidx(struct cxgbi_sock *csk, unsigned int tid,
+                               int pg_idx, bool reply)
+{
+       struct sk_buff *skb;
+       struct cpl_set_tcb_field *req;
+       u64 val = pg_idx < DDP_PGIDX_MAX ? pg_idx : 0;
+
+       if (!pg_idx)
+               return 0;
+
+       skb = alloc_cpl(sizeof(*req), 0, GFP_KERNEL);
+       if (!skb)
+               return -ENOMEM;
+
+       /*  set up ulp submode and page size */
+       val = (val & 0x03) << 2;
+       val |= TCB_ULP_TYPE(ULP2_MODE_ISCSI);
+
+       req = (struct cpl_set_tcb_field *)skb->head;
+       INIT_TP_WR(req, csk->tid);
+       OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, csk->tid));
+       req->reply_ctrl = htons(NO_REPLY(reply) | QUEUENO(csk->rss_qid));
+       req->word_cookie = htons(TCB_WORD(W_TCB_ULP_RAW));
+       req->mask = cpu_to_be64(TCB_ULP_TYPE(TCB_ULP_TYPE_MASK));
+       req->val = cpu_to_be64(val);
+       set_wr_txq(skb, CPL_PRIORITY_CONTROL, csk->port_id);
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p, tid 0x%x, pg_idx %u.\n", csk, csk->tid, pg_idx);
+
+       cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
+       return 0;
+}
+
+static int ddp_setup_conn_digest(struct cxgbi_sock *csk, unsigned int tid,
+                                int hcrc, int dcrc, int reply)
+{
+       struct sk_buff *skb;
+       struct cpl_set_tcb_field *req;
+       u64 val = (hcrc ? ULP_CRC_HEADER : 0) | (dcrc ? ULP_CRC_DATA : 0);
+
+       val = TCB_ULP_RAW(val);
+       val |= TCB_ULP_TYPE(ULP2_MODE_ISCSI);
+
+       skb = alloc_cpl(sizeof(*req), 0, GFP_KERNEL);
+       if (!skb)
+               return -ENOMEM;
+
+       csk->hcrc_len = (hcrc ? 4 : 0);
+       csk->dcrc_len = (dcrc ? 4 : 0);
+       /*  set up ulp submode and page size */
+       req = (struct cpl_set_tcb_field *)skb->head;
+       INIT_TP_WR(req, tid);
+       OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
+       req->reply_ctrl = htons(NO_REPLY(reply) | QUEUENO(csk->rss_qid));
+       req->word_cookie = htons(TCB_WORD(W_TCB_ULP_RAW));
+       req->mask = cpu_to_be64(TCB_ULP_RAW(TCB_ULP_RAW_MASK));
+       req->val = cpu_to_be64(val);
+       set_wr_txq(skb, CPL_PRIORITY_CONTROL, csk->port_id);
+
+       log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
+               "csk 0x%p, tid 0x%x, crc %d,%d.\n", csk, csk->tid, hcrc, dcrc);
+
+       cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
+       return 0;
+}
+
+static int cxgb4i_ddp_init(struct cxgbi_device *cdev)
+{
+       struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
+       struct cxgbi_ddp_info *ddp = cdev->ddp;
+       unsigned int tagmask, pgsz_factor[4];
+       int err;
+
+       if (ddp) {
+               kref_get(&ddp->refcnt);
+               pr_warn("cdev 0x%p, ddp 0x%p already set up.\n",
+                       cdev, cdev->ddp);
+               return -EALREADY;
+       }
+
+       err = cxgbi_ddp_init(cdev, lldi->vr->iscsi.start,
+                       lldi->vr->iscsi.start + lldi->vr->iscsi.size - 1,
+                       lldi->iscsi_iolen, lldi->iscsi_iolen);
+       if (err < 0)
+               return err;
+
+       ddp = cdev->ddp;
+
+       tagmask = ddp->idx_mask << PPOD_IDX_SHIFT;
+       cxgbi_ddp_page_size_factor(pgsz_factor);
+       cxgb4_iscsi_init(lldi->ports[0], tagmask, pgsz_factor);
+
+       cdev->csk_ddp_free_gl_skb = NULL;
+       cdev->csk_ddp_alloc_gl_skb = NULL;
+       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("cxgb4i 0x%p tag: sw %u, rsvd %u,%u, mask 0x%x.\n",
+               cdev, cdev->tag_format.sw_bits, cdev->tag_format.rsvd_bits,
+               cdev->tag_format.rsvd_shift, cdev->tag_format.rsvd_mask);
+       pr_info("cxgb4i 0x%p, nppods %u, bits %u, mask 0x%x,0x%x pkt %u/%u, "
+               " %u/%u.\n",
+               cdev, ddp->nppods, ddp->idx_bits, ddp->idx_mask,
+               ddp->rsvd_tag_mask, ddp->max_txsz, lldi->iscsi_iolen,
+               ddp->max_rxsz, lldi->iscsi_iolen);
+       pr_info("cxgb4i 0x%p max payload size: %u/%u, %u/%u.\n",
+               cdev, cdev->tx_max_size, ddp->max_txsz, cdev->rx_max_size,
+               ddp->max_rxsz);
+       return 0;
+}
+
+static void *t4_uld_add(const struct cxgb4_lld_info *lldi)
+{
+       struct cxgbi_device *cdev;
+       struct port_info *pi;
+       int i, rc;
+
+       cdev = cxgbi_device_register(sizeof(*lldi), lldi->nports);
+       if (!cdev) {
+               pr_info("t4 device 0x%p, register failed.\n", lldi);
+               return NULL;
+       }
+       pr_info("0x%p,0x%x, ports %u,%s, chan %u, q %u,%u, wr %u.\n",
+               cdev, lldi->adapter_type, lldi->nports,
+               lldi->ports[0]->name, lldi->nchan, lldi->ntxq,
+               lldi->nrxq, lldi->wr_cred);
+       for (i = 0; i < lldi->nrxq; i++)
+               log_debug(1 << CXGBI_DBG_DEV,
+                       "t4 0x%p, rxq id #%d: %u.\n",
+                       cdev, i, lldi->rxq_ids[i]);
+
+       memcpy(cxgbi_cdev_priv(cdev), lldi, sizeof(*lldi));
+       cdev->flags = CXGBI_FLAG_DEV_T4;
+       cdev->pdev = lldi->pdev;
+       cdev->ports = lldi->ports;
+       cdev->nports = lldi->nports;
+       cdev->mtus = lldi->mtus;
+       cdev->nmtus = NMTUS;
+       cdev->snd_win = cxgb4i_snd_win;
+       cdev->rcv_win = cxgb4i_rcv_win;
+       cdev->rx_credit_thres = cxgb4i_rx_credit_thres;
+       cdev->skb_tx_rsvd = CXGB4I_TX_HEADER_LEN;
+       cdev->skb_rx_extra = sizeof(struct cpl_iscsi_hdr);
+       cdev->itp = &cxgb4i_iscsi_transport;
+
+       rc = cxgb4i_ddp_init(cdev);
+       if (rc) {
+               pr_info("t4 0x%p ddp init failed.\n", cdev);
+               goto err_out;
+       }
+       rc = cxgb4i_ofld_init(cdev);
+       if (rc) {
+               pr_info("t4 0x%p ofld init failed.\n", cdev);
+               goto err_out;
+       }
+
+       rc = cxgbi_hbas_add(cdev, CXGB4I_MAX_LUN, CXGBI_MAX_CONN,
+                               &cxgb4i_host_template, cxgb4i_stt);
+       if (rc)
+               goto err_out;
+
+       for (i = 0; i < cdev->nports; i++) {
+               pi = netdev_priv(lldi->ports[i]);
+               cdev->hbas[i]->port_id = pi->port_id;
+       }
+       return cdev;
+
+err_out:
+       cxgbi_device_unregister(cdev);
+       return ERR_PTR(-ENOMEM);
+}
+
+#define RX_PULL_LEN    128
+static int t4_uld_rx_handler(void *handle, const __be64 *rsp,
+                               const struct pkt_gl *pgl)
+{
+       const struct cpl_act_establish *rpl;
+       struct sk_buff *skb;
+       unsigned int opc;
+       struct cxgbi_device *cdev = handle;
+
+       if (pgl == NULL) {
+               unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8;
+
+               skb = alloc_cpl(len, 0, GFP_ATOMIC);
+               if (!skb)
+                       goto nomem;
+               skb_copy_to_linear_data(skb, &rsp[1], len);
+       } else {
+               if (unlikely(*(u8 *)rsp != *(u8 *)pgl->va)) {
+                       pr_info("? FL 0x%p,RSS%#llx,FL %#llx,len %u.\n",
+                               pgl->va, be64_to_cpu(*rsp),
+                               be64_to_cpu(*(u64 *)pgl->va),
+                               pgl->tot_len);
+                       return 0;
+               }
+               skb = cxgb4_pktgl_to_skb(pgl, RX_PULL_LEN, RX_PULL_LEN);
+               if (unlikely(!skb))
+                       goto nomem;
+       }
+
+       rpl = (struct cpl_act_establish *)skb->data;
+       opc = rpl->ot.opcode;
+       log_debug(1 << CXGBI_DBG_TOE,
+               "cdev %p, opcode 0x%x(0x%x,0x%x), skb %p.\n",
+                cdev, opc, rpl->ot.opcode_tid, ntohl(rpl->ot.opcode_tid), skb);
+       if (cxgb4i_cplhandlers[opc])
+               cxgb4i_cplhandlers[opc](cdev, skb);
+       else {
+               pr_err("No handler for opcode 0x%x.\n", opc);
+               __kfree_skb(skb);
+       }
+       return 0;
+nomem:
+       log_debug(1 << CXGBI_DBG_TOE, "OOM bailing out.\n");
+       return 1;
+}
+
+static int t4_uld_state_change(void *handle, enum cxgb4_state state)
+{
+       struct cxgbi_device *cdev = handle;
+
+       switch (state) {
+       case CXGB4_STATE_UP:
+               pr_info("cdev 0x%p, UP.\n", cdev);
+               /* re-initialize */
+               break;
+       case CXGB4_STATE_START_RECOVERY:
+               pr_info("cdev 0x%p, RECOVERY.\n", cdev);
+               /* close all connections */
+               break;
+       case CXGB4_STATE_DOWN:
+               pr_info("cdev 0x%p, DOWN.\n", cdev);
+               break;
+       case CXGB4_STATE_DETACH:
+               pr_info("cdev 0x%p, DETACH.\n", cdev);
+               break;
+       default:
+               pr_info("cdev 0x%p, unknown state %d.\n", cdev, state);
+               break;
+       }
+       return 0;
+}
+
+static int __init cxgb4i_init_module(void)
+{
+       int rc;
+
+       printk(KERN_INFO "%s", version);
+
+       rc = cxgbi_iscsi_init(&cxgb4i_iscsi_transport, &cxgb4i_stt);
+       if (rc < 0)
+               return rc;
+       cxgb4_register_uld(CXGB4_ULD_ISCSI, &cxgb4i_uld_info);
+       return 0;
+}
+
+static void __exit cxgb4i_exit_module(void)
+{
+       cxgb4_unregister_uld(CXGB4_ULD_ISCSI);
+       cxgbi_device_unregister_all(CXGBI_FLAG_DEV_T4);
+       cxgbi_iscsi_cleanup(&cxgb4i_iscsi_transport, &cxgb4i_stt);
+}
+
+module_init(cxgb4i_init_module);
+module_exit(cxgb4i_exit_module);
diff --git a/drivers/scsi/cxgbi/cxgb4i/cxgb4i.h b/drivers/scsi/cxgbi/cxgb4i/cxgb4i.h
new file mode 100644 (file)
index 0000000..342263b
--- /dev/null
@@ -0,0 +1,38 @@
+/*
+ * cxgb4i.h: Chelsio T4 iSCSI driver.
+ *
+ * Copyright (c) 2010 Chelsio Communications, Inc.
+ *
+ * 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: Karen Xie (kxie@chelsio.com)
+ * Written by: Rakesh Ranjan (rranjan@chelsio.com)
+ */
+
+#ifndef        __CXGB4I_H__
+#define        __CXGB4I_H__
+
+#define        CXGB4I_SCSI_HOST_QDEPTH 1024
+#define        CXGB4I_MAX_CONN         16384
+#define        CXGB4I_MAX_TARGET       CXGB4I_MAX_CONN
+#define        CXGB4I_MAX_LUN          0x1000
+
+/* for TX: a skb must have a headroom of at least TX_HEADER_LEN bytes */
+#define CXGB4I_TX_HEADER_LEN \
+       (sizeof(struct fw_ofld_tx_data_wr) + sizeof(struct sge_opaque_hdr))
+
+struct cpl_rx_data_ddp {
+       union opcode_tid ot;
+       __be16 urg;
+       __be16 len;
+       __be32 seq;
+       union {
+               __be32 nxt_seq;
+               __be32 ddp_report;
+       };
+       __be32 ulp_crc;
+       __be32 ddpvld;
+};
+#endif /* __CXGB4I_H__ */