[SCSI] cxgb3i: Add cxgb3i iSCSI driver.

This patch implements the cxgb3i iscsi connection acceleration for the
open-iscsi initiator.

The cxgb3i 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>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

1 files changed, 1810 insertions, 0 deletions

diff --git a/drivers/scsi/cxgb3i/cxgb3i_offload.c b/drivers/scsi/cxgb3i/cxgb3i_offload.c
new file mode 100644
index 0000000..5f16081b
--- /dev/null
+++ b/drivers/scsi/cxgb3i/cxgb3i_offload.c
@@ -0,0 +1,1810 @@
+/*
+ * 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)
+ */
+
+#include <linux/if_vlan.h>
+#include <linux/version.h>
+
+#include "cxgb3_defs.h"
+#include "cxgb3_ctl_defs.h"
+#include "firmware_exports.h"
+#include "cxgb3i_offload.h"
+#include "cxgb3i_pdu.h"
+#include "cxgb3i_ddp.h"
+
+#ifdef __DEBUG_C3CN_CONN__
+#define c3cn_conn_debug         cxgb3i_log_info
+#else
+#define c3cn_conn_debug(fmt...)
+#endif
+
+#ifdef __DEBUG_C3CN_TX__
+#define c3cn_tx_debug         cxgb3i_log_debug
+#else
+#define c3cn_tx_debug(fmt...)
+#endif
+
+#ifdef __DEBUG_C3CN_RX__
+#define c3cn_rx_debug         cxgb3i_log_debug
+#else
+#define c3cn_rx_debug(fmt...)
+#endif
+
+/*
+ * module parameters releated to offloaded iscsi connection
+ */
+static int cxgb3_rcv_win = 256 * 1024;
+module_param(cxgb3_rcv_win, int, 0644);
+MODULE_PARM_DESC(cxgb3_rcv_win, "TCP receive window in bytes (default=256KB)");
+
+static int cxgb3_snd_win = 64 * 1024;
+module_param(cxgb3_snd_win, int, 0644);
+MODULE_PARM_DESC(cxgb3_snd_win, "TCP send window in bytes (default=64KB)");
+
+static int cxgb3_rx_credit_thres = 10 * 1024;
+module_param(cxgb3_rx_credit_thres, int, 0644);
+MODULE_PARM_DESC(rx_credit_thres,
+		 "RX credits return threshold in bytes (default=10KB)");
+
+static unsigned int cxgb3_max_connect = 8 * 1024;
+module_param(cxgb3_max_connect, uint, 0644);
+MODULE_PARM_DESC(cxgb3_max_connect, "Max. # of connections (default=8092)");
+
+static unsigned int cxgb3_sport_base = 20000;
+module_param(cxgb3_sport_base, uint, 0644);
+MODULE_PARM_DESC(cxgb3_sport_base, "starting port number (default=20000)");
+
+/*
+ * cxgb3i tcp connection data(per adapter) list
+ */
+static LIST_HEAD(cdata_list);
+static DEFINE_RWLOCK(cdata_rwlock);
+
+static int c3cn_push_tx_frames(struct s3_conn *c3cn, int req_completion);
+static void c3cn_release_offload_resources(struct s3_conn *c3cn);
+
+/*
+ * iscsi source port management
+ *
+ * Find a free source port in the port allocation map. We use a very simple
+ * rotor scheme to look for the next free port.
+ *
+ * If a source port has been specified make sure that it doesn't collide with
+ * our normal source port allocation map.  If it's outside the range of our
+ * allocation/deallocation scheme just let them use it.
+ *
+ * If the source port is outside our allocation range, the caller is
+ * responsible for keeping track of their port usage.
+ */
+static int c3cn_get_port(struct s3_conn *c3cn, struct cxgb3i_sdev_data *cdata)
+{
+	unsigned int start;
+	int idx;
+
+	if (!cdata)
+		goto error_out;
+
+	if (c3cn->saddr.sin_port != 0) {
+		idx = ntohs(c3cn->saddr.sin_port) - cxgb3_sport_base;
+		if (idx < 0 || idx >= cxgb3_max_connect)
+			return 0;
+		if (!test_and_set_bit(idx, cdata->sport_map))
+			return -EADDRINUSE;
+	}
+
+	/* the sport_map_next may not be accurate but that is okay, sport_map
+	   should be */
+	start = idx = cdata->sport_map_next;
+	do {
+		if (++idx >= cxgb3_max_connect)
+			idx = 0;
+		if (!(test_and_set_bit(idx, cdata->sport_map))) {
+			c3cn->saddr.sin_port = htons(cxgb3_sport_base + idx);
+			cdata->sport_map_next = idx;
+			c3cn_conn_debug("%s reserve port %u.\n",
+					cdata->cdev->name,
+					cxgb3_sport_base + idx);
+			return 0;
+		}
+	} while (idx != start);
+
+error_out:
+	return -EADDRNOTAVAIL;
+}
+
+static void c3cn_put_port(struct s3_conn *c3cn)
+{
+	struct cxgb3i_sdev_data *cdata = CXGB3_SDEV_DATA(c3cn->cdev);
+
+	if (c3cn->saddr.sin_port) {
+		int idx = ntohs(c3cn->saddr.sin_port) - cxgb3_sport_base;
+
+		c3cn->saddr.sin_port = 0;
+		if (idx < 0 || idx >= cxgb3_max_connect)
+			return;
+		clear_bit(idx, cdata->sport_map);
+		c3cn_conn_debug("%s, release port %u.\n",
+				cdata->cdev->name, cxgb3_sport_base + idx);
+	}
+}
+
+static inline void c3cn_set_flag(struct s3_conn *c3cn, enum c3cn_flags flag)
+{
+	__set_bit(flag, &c3cn->flags);
+	c3cn_conn_debug("c3cn 0x%p, set %d, s %u, f 0x%lx.\n",
+			c3cn, flag, c3cn->state, c3cn->flags);
+}
+
+static inline void c3cn_clear_flag(struct s3_conn *c3cn, enum c3cn_flags flag)
+{
+	__clear_bit(flag, &c3cn->flags);
+	c3cn_conn_debug("c3cn 0x%p, clear %d, s %u, f 0x%lx.\n",
+			c3cn, flag, c3cn->state, c3cn->flags);
+}
+
+static inline int c3cn_flag(struct s3_conn *c3cn, enum c3cn_flags flag)
+{
+	if (c3cn == NULL)
+		return 0;
+	return test_bit(flag, &c3cn->flags);
+}
+
+static void c3cn_set_state(struct s3_conn *c3cn, int state)
+{
+	c3cn_conn_debug("c3cn 0x%p state -> %u.\n", c3cn, state);
+	c3cn->state = state;
+}
+
+static inline void c3cn_hold(struct s3_conn *c3cn)
+{
+	atomic_inc(&c3cn->refcnt);
+}
+
+static inline void c3cn_put(struct s3_conn *c3cn)
+{
+	if (atomic_dec_and_test(&c3cn->refcnt)) {
+		c3cn_conn_debug("free c3cn 0x%p, s %u, f 0x%lx.\n",
+				c3cn, c3cn->state, c3cn->flags);
+		kfree(c3cn);
+	}
+}
+
+static void c3cn_closed(struct s3_conn *c3cn)
+{
+	c3cn_conn_debug("c3cn 0x%p, state %u, flag 0x%lx.\n",
+			 c3cn, c3cn->state, c3cn->flags);
+
+	c3cn_put_port(c3cn);
+	c3cn_release_offload_resources(c3cn);
+	c3cn_set_state(c3cn, C3CN_STATE_CLOSED);
+	cxgb3i_conn_closing(c3cn);
+}
+
+/*
+ * CPL (Chelsio Protocol Language) defines a message passing interface between
+ * the host driver and T3 asic.
+ * The section below implments CPLs that related to iscsi tcp connection
+ * open/close/abort and data send/receive.
+ */
+
+/*
+ * CPL connection active open request: host ->
+ */
+static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
+{
+	int i = 0;
+
+	while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
+		++i;
+	return i;
+}
+
+static unsigned int select_mss(struct s3_conn *c3cn, unsigned int pmtu)
+{
+	unsigned int idx;
+	struct dst_entry *dst = c3cn->dst_cache;
+	struct t3cdev *cdev = c3cn->cdev;
+	const struct t3c_data *td = T3C_DATA(cdev);
+	u16 advmss = dst_metric(dst, RTAX_ADVMSS);
+
+	if (advmss > pmtu - 40)
+		advmss = pmtu - 40;
+	if (advmss < td->mtus[0] - 40)
+		advmss = td->mtus[0] - 40;
+	idx = find_best_mtu(td, advmss + 40);
+	return idx;
+}
+
+static inline int compute_wscale(int win)
+{
+	int wscale = 0;
+	while (wscale < 14 && (65535<<wscale) < win)
+		wscale++;
+	return wscale;
+}
+
+static inline unsigned int calc_opt0h(struct s3_conn *c3cn)
+{
+	int wscale = compute_wscale(cxgb3_rcv_win);
+	return  V_KEEP_ALIVE(1) |
+		F_TCAM_BYPASS |
+		V_WND_SCALE(wscale) |
+		V_MSS_IDX(c3cn->mss_idx);
+}
+
+static inline unsigned int calc_opt0l(struct s3_conn *c3cn)
+{
+	return  V_ULP_MODE(ULP_MODE_ISCSI) |
+		V_RCV_BUFSIZ(cxgb3_rcv_win>>10);
+}
+
+static void make_act_open_req(struct s3_conn *c3cn, struct sk_buff *skb,
+			      unsigned int atid, const struct l2t_entry *e)
+{
+	struct cpl_act_open_req *req;
+
+	c3cn_conn_debug("c3cn 0x%p, atid 0x%x.\n", c3cn, atid);
+
+	skb->priority = CPL_PRIORITY_SETUP;
+	req = (struct cpl_act_open_req *)__skb_put(skb, sizeof(*req));
+	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
+	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, atid));
+	req->local_port = c3cn->saddr.sin_port;
+	req->peer_port = c3cn->daddr.sin_port;
+	req->local_ip = c3cn->saddr.sin_addr.s_addr;
+	req->peer_ip = c3cn->daddr.sin_addr.s_addr;
+	req->opt0h = htonl(calc_opt0h(c3cn) | V_L2T_IDX(e->idx) |
+			   V_TX_CHANNEL(e->smt_idx));
+	req->opt0l = htonl(calc_opt0l(c3cn));
+	req->params = 0;
+}
+
+static void fail_act_open(struct s3_conn *c3cn, int errno)
+{
+	c3cn_conn_debug("c3cn 0x%p, state %u, flag 0x%lx.\n",
+			c3cn, c3cn->state, c3cn->flags);
+	c3cn->err = errno;
+	c3cn_closed(c3cn);
+}
+
+static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
+{
+	struct s3_conn *c3cn = (struct s3_conn *)skb->sk;
+
+	c3cn_conn_debug("c3cn 0x%p, state %u.\n", c3cn, c3cn->state);
+
+	c3cn_hold(c3cn);
+	spin_lock_bh(&c3cn->lock);
+	if (c3cn->state == C3CN_STATE_CONNECTING)
+		fail_act_open(c3cn, EHOSTUNREACH);
+	spin_unlock_bh(&c3cn->lock);
+	c3cn_put(c3cn);
+	__kfree_skb(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 skb_entail(struct s3_conn *c3cn, struct sk_buff *skb,
+		       int flags)
+{
+	CXGB3_SKB_CB(skb)->seq = c3cn->write_seq;
+	CXGB3_SKB_CB(skb)->flags = flags;
+	__skb_queue_tail(&c3cn->write_queue, skb);
+}
+
+static void send_close_req(struct s3_conn *c3cn)
+{
+	struct sk_buff *skb = c3cn->cpl_close;
+	struct cpl_close_con_req *req = (struct cpl_close_con_req *)skb->head;
+	unsigned int tid = c3cn->tid;
+
+	c3cn_conn_debug("c3cn 0x%p, state 0x%x, flag 0x%lx.\n",
+			c3cn, c3cn->state, c3cn->flags);
+
+	c3cn->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(c3cn->write_seq);
+
+	skb_entail(c3cn, skb, C3CB_FLAG_NO_APPEND);
+	if (c3cn->state != C3CN_STATE_CONNECTING)
+		c3cn_push_tx_frames(c3cn, 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 *cdev, struct sk_buff *skb)
+{
+	struct cpl_abort_req *req = cplhdr(skb);
+
+	c3cn_conn_debug("tdev 0x%p.\n", cdev);
+
+	req->cmd = CPL_ABORT_NO_RST;
+	cxgb3_ofld_send(cdev, skb);
+}
+
+static inline void c3cn_purge_write_queue(struct s3_conn *c3cn)
+{
+	struct sk_buff *skb;
+
+	while ((skb = __skb_dequeue(&c3cn->write_queue)))
+		__kfree_skb(skb);
+}
+
+static void send_abort_req(struct s3_conn *c3cn)
+{
+	struct sk_buff *skb = c3cn->cpl_abort_req;
+	struct cpl_abort_req *req;
+	unsigned int tid = c3cn->tid;
+
+	if (unlikely(c3cn->state == C3CN_STATE_ABORTING) || !skb ||
+		     !c3cn->cdev)
+		return;
+
+	c3cn_set_state(c3cn, C3CN_STATE_ABORTING);
+
+	c3cn_conn_debug("c3cn 0x%p, flag ABORT_RPL + ABORT_SHUT.\n", c3cn);
+
+	c3cn_set_flag(c3cn, C3CN_ABORT_RPL_PENDING);
+
+	/* Purge the send queue so we don't send anything after an abort. */
+	c3cn_purge_write_queue(c3cn);
+
+	c3cn->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(tid));
+	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, tid));
+	req->rsvd0 = htonl(c3cn->snd_nxt);
+	req->rsvd1 = !c3cn_flag(c3cn, C3CN_TX_DATA_SENT);
+	req->cmd = CPL_ABORT_SEND_RST;
+
+	l2t_send(c3cn->cdev, skb, c3cn->l2t);
+}
+
+/*
+ * CPL connection abort reply: host ->
+ *
+ * Send an ABORT_RPL message in response of the ABORT_REQ received.
+ */
+static void send_abort_rpl(struct s3_conn *c3cn, int rst_status)
+{
+	struct sk_buff *skb = c3cn->cpl_abort_rpl;
+	struct cpl_abort_rpl *rpl = (struct cpl_abort_rpl *)skb->head;
+
+	c3cn->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(c3cn->tid));
+	OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, c3cn->tid));
+	rpl->cmd = rst_status;
+
+	cxgb3_ofld_send(c3cn->cdev, 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 s3_conn *c3cn, u32 credits, u32 dack)
+{
+	struct sk_buff *skb;
+	struct cpl_rx_data_ack *req;
+
+	skb = alloc_skb(sizeof(*req), GFP_ATOMIC);
+	if (!skb)
+		return 0;
+
+	req = (struct cpl_rx_data_ack *)__skb_put(skb, sizeof(*req));
+	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
+	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, c3cn->tid));
+	req->credit_dack = htonl(dack | V_RX_CREDITS(credits));
+	skb->priority = CPL_PRIORITY_ACK;
+	cxgb3_ofld_send(c3cn->cdev, 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).
+ */
+
+/*
+ * For ULP connections HW may inserts digest bytes into the pdu. Those digest
+ * bytes are not sent by the host but are part of the TCP payload and therefore
+ * consume TCP sequence space.
+ */
+static const unsigned int cxgb3_ulp_extra_len[] = { 0, 4, 4, 8 };
+static inline unsigned int ulp_extra_len(const struct sk_buff *skb)
+{
+	return cxgb3_ulp_extra_len[skb_ulp_mode(skb) & 3];
+}
+
+static unsigned int wrlen __read_mostly;
+
+/*
+ * The number of WRs needed for an skb depends on the number of fragments
+ * in the skb and whether it has any payload in its main body.  This maps the
+ * length of the gather list represented by an skb into the # of necessary WRs.
+ *
+ * The max. length of an skb is controlled by the max pdu size which is ~16K.
+ * Also, assume the min. fragment length is the sector size (512), then add
+ * extra fragment counts for iscsi bhs and payload padding.
+ */
+#define SKB_WR_LIST_SIZE	(16384/512 + 3)
+static unsigned int skb_wrs[SKB_WR_LIST_SIZE] __read_mostly;
+
+static void s3_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 reset_wr_list(struct s3_conn *c3cn)
+{
+	c3cn->wr_pending_head = NULL;
+}
+
+/*
+ * Add a WR to a connections's list of pending WRs.  This is a singly-linked
+ * list of sk_buffs operating as a FIFO.  The head is kept in wr_pending_head
+ * and the tail in wr_pending_tail.
+ */
+static inline void enqueue_wr(struct s3_conn *c3cn,
+			      struct sk_buff *skb)
+{
+	skb->sp = NULL;
+
+	/*
+	 * We want to take an extra reference since both us and the driver
+	 * need to free the packet before it's really freed. We know there's
+	 * just one user currently so we use atomic_set rather than skb_get
+	 * to avoid the atomic op.
+	 */
+	atomic_set(&skb->users, 2);
+
+	if (!c3cn->wr_pending_head)
+		c3cn->wr_pending_head = skb;
+	else
+		c3cn->wr_pending_tail->sp = (void *)skb;
+	c3cn->wr_pending_tail = skb;
+}
+
+static inline struct sk_buff *peek_wr(const struct s3_conn *c3cn)
+{
+	return c3cn->wr_pending_head;
+}
+
+static inline void free_wr_skb(struct sk_buff *skb)
+{
+	kfree_skb(skb);
+}
+
+static inline struct sk_buff *dequeue_wr(struct s3_conn *c3cn)
+{
+	struct sk_buff *skb = c3cn->wr_pending_head;
+
+	if (likely(skb)) {
+		/* Don't bother clearing the tail */
+		c3cn->wr_pending_head = (struct sk_buff *)skb->sp;
+		skb->sp = NULL;
+	}
+	return skb;
+}
+
+static void purge_wr_queue(struct s3_conn *c3cn)
+{
+	struct sk_buff *skb;
+	while ((skb = dequeue_wr(c3cn)) != NULL)
+		free_wr_skb(skb);
+}
+
+static inline void make_tx_data_wr(struct s3_conn *c3cn, struct sk_buff *skb,
+				   int len)
+{
+	struct tx_data_wr *req;
+
+	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->wr_lo = htonl(V_WR_TID(c3cn->tid));
+	req->sndseq = htonl(c3cn->snd_nxt);
+	/* len includes the length of any HW ULP additions */
+	req->len = htonl(len);
+	req->param = htonl(V_TX_PORT(c3cn->l2t->smt_idx));
+	/* V_TX_ULP_SUBMODE sets both the mode and submode */
+	req->flags = htonl(V_TX_ULP_SUBMODE(skb_ulp_mode(skb)) |
+			   V_TX_SHOVE((skb_peek(&c3cn->write_queue) ? 0 : 1)));
+
+	if (!c3cn_flag(c3cn, C3CN_TX_DATA_SENT)) {
+		req->flags |= htonl(V_TX_ACK_PAGES(2) | F_TX_INIT |
+				    V_TX_CPU_IDX(c3cn->qset));
+		/* Sendbuffer is in units of 32KB. */
+		req->param |= htonl(V_TX_SNDBUF(cxgb3_snd_win >> 15));
+		c3cn_set_flag(c3cn, C3CN_TX_DATA_SENT);
+	}
+}
+
+/**
+ * c3cn_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_discard(struct t3cdev *cdev, struct sk_buff *skb)
+{
+	kfree_skb(skb);
+}
+
+static int c3cn_push_tx_frames(struct s3_conn *c3cn, int req_completion)
+{
+	int total_size = 0;
+	struct sk_buff *skb;
+	struct t3cdev *cdev;
+	struct cxgb3i_sdev_data *cdata;
+
+	if (unlikely(c3cn->state == C3CN_STATE_CONNECTING ||
+		     c3cn->state == C3CN_STATE_CLOSE_WAIT_1 ||
+		     c3cn->state == C3CN_STATE_ABORTING)) {
+		c3cn_tx_debug("c3cn 0x%p, in closing state %u.\n",
+			      c3cn, c3cn->state);
+		return 0;
+	}
+
+	cdev = c3cn->cdev;
+	cdata = CXGB3_SDEV_DATA(cdev);
+
+	while (c3cn->wr_avail
+	       && (skb = skb_peek(&c3cn->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 (c3cn->wr_avail < wrs_needed) {
+			c3cn_tx_debug("c3cn 0x%p, skb len %u/%u, frag %u, "
+				      "wr %d < %u.\n",
+				      c3cn, skb->len, skb->datalen, frags,
+				      wrs_needed, c3cn->wr_avail);
+			break;
+		}
+
+		__skb_unlink(skb, &c3cn->write_queue);
+		skb->priority = CPL_PRIORITY_DATA;
+		skb->csum = wrs_needed;	/* remember this until the WR_ACK */
+		c3cn->wr_avail -= wrs_needed;
+		c3cn->wr_unacked += wrs_needed;
+		enqueue_wr(c3cn, skb);
+
+		if (likely(CXGB3_SKB_CB(skb)->flags & C3CB_FLAG_NEED_HDR)) {
+			len += ulp_extra_len(skb);
+			make_tx_data_wr(c3cn, skb, len);
+			c3cn->snd_nxt += len;
+			if ((req_completion
+			     && c3cn->wr_unacked == wrs_needed)
+			    || (CXGB3_SKB_CB(skb)->flags & C3CB_FLAG_COMPL)
+			    || c3cn->wr_unacked >= c3cn->wr_max / 2) {
+				struct work_request_hdr *wr = cplhdr(skb);
+
+				wr->wr_hi |= htonl(F_WR_COMPL);
+				c3cn->wr_unacked = 0;
+			}
+			CXGB3_SKB_CB(skb)->flags &= ~C3CB_FLAG_NEED_HDR;
+		}
+
+		total_size += skb->truesize;
+		set_arp_failure_handler(skb, arp_failure_discard);
+		l2t_send(cdev, skb, c3cn->l2t);
+	}
+	return total_size;
+}
+
+/*
+ * process_cpl_msg: -> host
+ * Top-level CPL message processing used by most CPL messages that
+ * pertain to connections.
+ */
+static inline void process_cpl_msg(void (*fn)(struct s3_conn *,
+					      struct sk_buff *),
+				   struct s3_conn *c3cn,
+				   struct sk_buff *skb)
+{
+	spin_lock_bh(&c3cn->lock);
+	fn(c3cn, skb);
+	spin_unlock_bh(&c3cn->lock);
+}
+
+/*
+ * process_cpl_msg_ref: -> host
+ * Similar to process_cpl_msg() but takes an extra connection reference around
+ * the call to the handler.  Should be used if the handler may drop a
+ * connection reference.
+ */
+static inline void process_cpl_msg_ref(void (*fn) (struct s3_conn *,
+						   struct sk_buff *),
+				       struct s3_conn *c3cn,
+				       struct sk_buff *skb)
+{
+	c3cn_hold(c3cn);
+	process_cpl_msg(fn, c3cn, skb);
+	c3cn_put(c3cn);
+}
+
+/*
+ * 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 s3_free_atid(struct t3cdev *cdev, unsigned int tid)
+{
+	struct s3_conn *c3cn = cxgb3_free_atid(cdev, tid);
+	if (c3cn)
+		c3cn_put(c3cn);
+}
+
+static void c3cn_established(struct s3_conn *c3cn, u32 snd_isn,
+			     unsigned int opt)
+{
+	c3cn_conn_debug("c3cn 0x%p, state %u.\n", c3cn, c3cn->state);
+
+	c3cn->write_seq = c3cn->snd_nxt = c3cn->snd_una = snd_isn;
+
+	/*
+	 * Causes the first RX_DATA_ACK to supply any Rx credits we couldn't
+	 * pass through opt0.
+	 */
+	if (cxgb3_rcv_win > (M_RCV_BUFSIZ << 10))
+		c3cn->rcv_wup -= cxgb3_rcv_win - (M_RCV_BUFSIZ << 10);
+
+	dst_confirm(c3cn->dst_cache);
+
+	smp_mb();
+
+	c3cn_set_state(c3cn, C3CN_STATE_ESTABLISHED);
+}
+
+static void process_act_establish(struct s3_conn *c3cn, struct sk_buff *skb)
+{
+	struct cpl_act_establish *req = cplhdr(skb);
+	u32 rcv_isn = ntohl(req->rcv_isn);	/* real RCV_ISN + 1 */
+
+	c3cn_conn_debug("c3cn 0x%p, state %u, flag 0x%lx.\n",
+			c3cn, c3cn->state, c3cn->flags);
+
+	if (unlikely(c3cn->state != C3CN_STATE_CONNECTING))
+		cxgb3i_log_error("TID %u expected SYN_SENT, got EST., s %u\n",
+				 c3cn->tid, c3cn->state);
+
+	c3cn->copied_seq = c3cn->rcv_wup = c3cn->rcv_nxt = rcv_isn;
+	c3cn_established(c3cn, ntohl(req->snd_isn), ntohs(req->tcp_opt));
+
+	__kfree_skb(skb);
+
+	if (unlikely(c3cn_flag(c3cn, C3CN_ACTIVE_CLOSE_NEEDED)))
+		/* upper layer has requested closing */
+		send_abort_req(c3cn);
+	else if (c3cn_push_tx_frames(c3cn, 1))
+		cxgb3i_conn_tx_open(c3cn);
+}
+
+static int do_act_establish(struct t3cdev *cdev, struct sk_buff *skb,
+			    void *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));
+	struct s3_conn *c3cn = ctx;
+	struct cxgb3i_sdev_data *cdata = CXGB3_SDEV_DATA(cdev);
+
+	c3cn_conn_debug("rcv, tid 0x%x, c3cn 0x%p, s %u, f 0x%lx.\n",
+			tid, c3cn, c3cn->state, c3cn->flags);
+
+	c3cn->tid = tid;
+	c3cn_hold(c3cn);
+	cxgb3_insert_tid(cdata->cdev, cdata->client, c3cn, tid);
+	s3_free_atid(cdev, atid);
+
+	c3cn->qset = G_QNUM(ntohl(skb->csum));
+
+	process_cpl_msg(process_act_establish, c3cn, 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:
+		cxgb3i_log_error("ACTIVE_OPEN_RPL: 4-tuple in use\n");
+		return EADDRINUSE;
+	default:
+		return EIO;
+	}
+}
+
+static void act_open_retry_timer(unsigned long data)
+{
+	struct sk_buff *skb;
+	struct s3_conn *c3cn = (struct s3_conn *)data;
+
+	c3cn_conn_debug("c3cn 0x%p, state %u.\n", c3cn, c3cn->state);
+
+	spin_lock_bh(&c3cn->lock);
+	skb = alloc_skb(sizeof(struct cpl_act_open_req), GFP_ATOMIC);
+	if (!skb)
+		fail_act_open(c3cn, ENOMEM);
+	else {
+		skb->sk = (struct sock *)c3cn;
+		set_arp_failure_handler(skb, act_open_req_arp_failure);
+		make_act_open_req(c3cn, skb, c3cn->tid, c3cn->l2t);
+		l2t_send(c3cn->cdev, skb, c3cn->l2t);
+	}
+	spin_unlock_bh(&c3cn->lock);
+	c3cn_put(c3cn);
+}
+
+static void process_act_open_rpl(struct s3_conn *c3cn, struct sk_buff *skb)
+{
+	struct cpl_act_open_rpl *rpl = cplhdr(skb);
+
+	c3cn_conn_debug("c3cn 0x%p, state %u, flag 0x%lx.\n",
+			c3cn, c3cn->state, c3cn->flags);
+
+	if (rpl->status == CPL_ERR_CONN_EXIST &&
+	    c3cn->retry_timer.function != act_open_retry_timer) {
+		c3cn->retry_timer.function = act_open_retry_timer;
+		if (!mod_timer(&c3cn->retry_timer, jiffies + HZ / 2))
+			c3cn_hold(c3cn);
+	} else
+		fail_act_open(c3cn, act_open_rpl_status_to_errno(rpl->status));
+	__kfree_skb(skb);
+}
+
+static int do_act_open_rpl(struct t3cdev *cdev, struct sk_buff *skb, void *ctx)
+{
+	struct s3_conn *c3cn = ctx;
+	struct cpl_act_open_rpl *rpl = cplhdr(skb);
+
+	c3cn_conn_debug("rcv, status 0x%x, c3cn 0x%p, s %u, f 0x%lx.\n",
+			rpl->status, c3cn, c3cn->state, c3cn->flags);
+
+	if (rpl->status != CPL_ERR_TCAM_FULL &&
+	    rpl->status != CPL_ERR_CONN_EXIST &&
+	    rpl->status != CPL_ERR_ARP_MISS)
+		cxgb3_queue_tid_release(cdev, GET_TID(rpl));
+
+	process_cpl_msg_ref(process_act_open_rpl, c3cn, skb);
+	return 0;
+}
+
+/*
+ * Process PEER_CLOSE CPL messages: -> host
+ * Handle peer FIN.
+ */
+static void process_peer_close(struct s3_conn *c3cn, struct sk_buff *skb)
+{
+	c3cn_conn_debug("c3cn 0x%p, state %u, flag 0x%lx.\n",
+			c3cn, c3cn->state, c3cn->flags);
+
+	if (c3cn_flag(c3cn, C3CN_ABORT_RPL_PENDING))
+		goto out;
+
+	switch (c3cn->state) {
+	case C3CN_STATE_ESTABLISHED:
+		c3cn_set_state(c3cn, C3CN_STATE_PASSIVE_CLOSE);
+		break;
+	case C3CN_STATE_ACTIVE_CLOSE:
+		c3cn_set_state(c3cn, C3CN_STATE_CLOSE_WAIT_2);
+		break;
+	case C3CN_STATE_CLOSE_WAIT_1:
+		c3cn_closed(c3cn);
+		break;
+	case C3CN_STATE_ABORTING:
+		break;
+	default:
+		cxgb3i_log_error("%s: peer close, TID %u in bad state %u\n",
+				 c3cn->cdev->name, c3cn->tid, c3cn->state);
+	}
+
+	cxgb3i_conn_closing(c3cn);
+out:
+	__kfree_skb(skb);
+}
+
+static int do_peer_close(struct t3cdev *cdev, struct sk_buff *skb, void *ctx)
+{
+	struct s3_conn *c3cn = ctx;
+
+	c3cn_conn_debug("rcv, c3cn 0x%p, s %u, f 0x%lx.\n",
+			c3cn, c3cn->state, c3cn->flags);
+	process_cpl_msg_ref(process_peer_close, c3cn, skb);
+	return 0;
+}
+
+/*
+ * Process CLOSE_CONN_RPL CPL message: -> host
+ * Process a peer ACK to our FIN.
+ */
+static void process_close_con_rpl(struct s3_conn *c3cn, struct sk_buff *skb)
+{
+	struct cpl_close_con_rpl *rpl = cplhdr(skb);
+
+	c3cn_conn_debug("c3cn 0x%p, state %u, flag 0x%lx.\n",
+			c3cn, c3cn->state, c3cn->flags);
+
+	c3cn->snd_una = ntohl(rpl->snd_nxt) - 1;	/* exclude FIN */
+
+	if (c3cn_flag(c3cn, C3CN_ABORT_RPL_PENDING))
+		goto out;
+
+	switch (c3cn->state) {
+	case C3CN_STATE_ACTIVE_CLOSE:
+		c3cn_set_state(c3cn, C3CN_STATE_CLOSE_WAIT_1);
+		break;
+	case C3CN_STATE_CLOSE_WAIT_1:
+	case C3CN_STATE_CLOSE_WAIT_2:
+		c3cn_closed(c3cn);
+		break;
+	case C3CN_STATE_ABORTING:
+		break;
+	default:
+		cxgb3i_log_error("%s: close_rpl, TID %u in bad state %u\n",
+				 c3cn->cdev->name, c3cn->tid, c3cn->state);
+	}
+
+out:
+	kfree_skb(skb);
+}
+
+static int do_close_con_rpl(struct t3cdev *cdev, struct sk_buff *skb,
+			    void *ctx)
+{
+	struct s3_conn *c3cn = ctx;
+
+	c3cn_conn_debug("rcv, c3cn 0x%p, s %u, f 0x%lx.\n",
+			 c3cn, c3cn->state, c3cn->flags);
+
+	process_cpl_msg_ref(process_close_con_rpl, c3cn, 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 s3_conn *c3cn, int abort_reason,
+				 int *need_rst)
+{
+	switch (abort_reason) {
+	case CPL_ERR_BAD_SYN: /* fall through */
+	case CPL_ERR_CONN_RESET:
+		return c3cn->state > C3CN_STATE_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 process_abort_req(struct s3_conn *c3cn, struct sk_buff *skb)
+{
+	int rst_status = CPL_ABORT_NO_RST;
+	const struct cpl_abort_req_rss *req = cplhdr(skb);
+
+	c3cn_conn_debug("c3cn 0x%p, state %u, flag 0x%lx.\n",
+			c3cn, c3cn->state, c3cn->flags);
+
+	if (!c3cn_flag(c3cn, C3CN_ABORT_REQ_RCVD)) {
+		c3cn_set_flag(c3cn, C3CN_ABORT_REQ_RCVD);
+		c3cn_set_state(c3cn, C3CN_STATE_ABORTING);
+		__kfree_skb(skb);
+		return;
+	}
+
+	c3cn_clear_flag(c3cn, C3CN_ABORT_REQ_RCVD);
+	send_abort_rpl(c3cn, rst_status);
+
+	if (!c3cn_flag(c3cn, C3CN_ABORT_RPL_PENDING)) {
+		c3cn->err =
+		    abort_status_to_errno(c3cn, req->status, &rst_status);
+		c3cn_closed(c3cn);
+	}
+}
+
+static int do_abort_req(struct t3cdev *cdev, struct sk_buff *skb, void *ctx)
+{
+	const struct cpl_abort_req_rss *req = cplhdr(skb);
+	struct s3_conn *c3cn = ctx;
+
+	c3cn_conn_debug("rcv, c3cn 0x%p, s 0x%x, f 0x%lx.\n",
+			c3cn, c3cn->state, c3cn->flags);
+
+	if (req->status == CPL_ERR_RTX_NEG_ADVICE ||
+	    req->status == CPL_ERR_PERSIST_NEG_ADVICE) {
+		__kfree_skb(skb);
+		return 0;
+	}
+
+	process_cpl_msg_ref(process_abort_req, c3cn, 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 void process_abort_rpl(struct s3_conn *c3cn, struct sk_buff *skb)
+{
+	c3cn_conn_debug("c3cn 0x%p, state %u, flag 0x%lx.\n",
+			c3cn, c3cn->state, c3cn->flags);
+
+	if (c3cn_flag(c3cn, C3CN_ABORT_RPL_PENDING)) {
+		if (!c3cn_flag(c3cn, C3CN_ABORT_RPL_RCVD))
+			c3cn_set_flag(c3cn, C3CN_ABORT_RPL_RCVD);
+		else {
+			c3cn_clear_flag(c3cn, C3CN_ABORT_RPL_RCVD);
+			c3cn_clear_flag(c3cn, C3CN_ABORT_RPL_PENDING);
+			if (c3cn_flag(c3cn, C3CN_ABORT_REQ_RCVD))
+				cxgb3i_log_error("%s tid %u, ABORT_RPL_RSS\n",
+						 c3cn->cdev->name, c3cn->tid);
+			c3cn_closed(c3cn);
+		}
+	}
+	__kfree_skb(skb);
+}
+
+static int do_abort_rpl(struct t3cdev *cdev, struct sk_buff *skb, void *ctx)
+{
+	struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
+	struct s3_conn *c3cn = ctx;
+
+	c3cn_conn_debug("rcv, status 0x%x, c3cn 0x%p, s %u, 0x%lx.\n",
+			rpl->status, c3cn, c3cn ? c3cn->state : 0,
+			c3cn ? c3cn->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 discard;
+
+	/*
+	 * 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 (!c3cn)
+		goto discard;
+
+	process_cpl_msg_ref(process_abort_rpl, c3cn, skb);
+	return 0;
+
+discard:
+	__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 void process_rx_iscsi_hdr(struct s3_conn *c3cn, struct sk_buff *skb)
+{
+	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;
+
+	if (unlikely(c3cn->state >= C3CN_STATE_PASSIVE_CLOSE)) {
+		if (c3cn->state != C3CN_STATE_ABORTING)
+			send_abort_req(c3cn);
+		__kfree_skb(skb);
+		return;
+	}
+
+	CXGB3_SKB_CB(skb)->seq = ntohl(hdr_cpl->seq);
+	CXGB3_SKB_CB(skb)->flags = 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) {
+		cxgb3i_log_error("%s: TID %u, ISCSI_HDR, skb len %u < %u.\n",
+				 c3cn->cdev->name, c3cn->tid,
+				 skb->len, hdr_len);
+		goto abort_conn;
+	}
+
+	err = skb_copy_bits(skb, skb->len - sizeof(ddp_cpl), &ddp_cpl,
+			    sizeof(ddp_cpl));
+	if (err < 0)
+		goto abort_conn;
+
+	skb_ulp_mode(skb) = ULP2_FLAG_DATA_READY;
+	skb_ulp_pdulen(skb) = ntohs(ddp_cpl.len);
+	skb_ulp_ddigest(skb) = ntohl(ddp_cpl.ulp_crc);
+	status = ntohl(ddp_cpl.ddp_status);
+
+	c3cn_rx_debug("rx skb 0x%p, len %u, pdulen %u, ddp status 0x%x.\n",
+		      skb, skb->len, skb_ulp_pdulen(skb), status);
+
+	if (status & (1 << RX_DDP_STATUS_HCRC_SHIFT))
+		skb_ulp_mode(skb) |= ULP2_FLAG_HCRC_ERROR;
+	if (status & (1 << RX_DDP_STATUS_DCRC_SHIFT))
+		skb_ulp_mode(skb) |= ULP2_FLAG_DCRC_ERROR;
+	if (status & (1 << RX_DDP_STATUS_PAD_SHIFT))
+		skb_ulp_mode(skb) |= ULP2_FLAG_PAD_ERROR;
+
+	if (skb->len > (hdr_len + sizeof(ddp_cpl))) {
+		err = skb_copy_bits(skb, hdr_len, &data_cpl, sizeof(data_cpl));
+		if (err < 0)
+			goto abort_conn;
+		data_len = ntohs(data_cpl.len);
+		len += sizeof(data_cpl) + data_len;
+	} else if (status & (1 << RX_DDP_STATUS_DDP_SHIFT))
+		skb_ulp_mode(skb) |= ULP2_FLAG_DATA_DDPED;
+
+	c3cn->rcv_nxt = ntohl(ddp_cpl.seq) + skb_ulp_pdulen(skb);
+	__pskb_trim(skb, len);
+	__skb_queue_tail(&c3cn->receive_queue, skb);
+	cxgb3i_conn_pdu_ready(c3cn);
+
+	return;
+
+abort_conn:
+	send_abort_req(c3cn);
+	__kfree_skb(skb);
+}
+
+static int do_iscsi_hdr(struct t3cdev *t3dev, struct sk_buff *skb, void *ctx)
+{
+	struct s3_conn *c3cn = ctx;
+
+	process_cpl_msg(process_rx_iscsi_hdr, c3cn, 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 void process_wr_ack(struct s3_conn *c3cn, struct sk_buff *skb)
+{
+	struct cpl_wr_ack *hdr = cplhdr(skb);
+	unsigned int credits = ntohs(hdr->credits);
+	u32 snd_una = ntohl(hdr->snd_una);
+
+	c3cn->wr_avail += credits;
+	if (c3cn->wr_unacked > c3cn->wr_max - c3cn->wr_avail)
+		c3cn->wr_unacked = c3cn->wr_max - c3cn->wr_avail;
+
+	while (credits) {
+		struct sk_buff *p = peek_wr(c3cn);
+
+		if (unlikely(!p)) {
+			cxgb3i_log_error("%u WR_ACK credits for TID %u with "
+					 "nothing pending, state %u\n",
+					 credits, c3cn->tid, c3cn->state);
+			break;
+		}
+		if (unlikely(credits < p->csum)) {
+			p->csum -= credits;
+			break;
+		} else {
+			dequeue_wr(c3cn);
+			credits -= p->csum;
+			free_wr_skb(p);
+		}
+	}
+
+	if (unlikely(before(snd_una, c3cn->snd_una)))
+		goto out_free;
+
+	if (c3cn->snd_una != snd_una) {
+		c3cn->snd_una = snd_una;
+		dst_confirm(c3cn->dst_cache);
+	}
+
+	if (skb_queue_len(&c3cn->write_queue) && c3cn_push_tx_frames(c3cn, 0))
+		cxgb3i_conn_tx_open(c3cn);
+out_free:
+	__kfree_skb(skb);
+}
+
+static int do_wr_ack(struct t3cdev *cdev, struct sk_buff *skb, void *ctx)
+{
+	struct s3_conn *c3cn = ctx;
+
+	process_cpl_msg(process_wr_ack, c3cn, skb);
+	return 0;
+}
+
+/*
+ * for each connection, pre-allocate skbs needed for close/abort requests. So
+ * that we can service the request right away.
+ */
+static void c3cn_free_cpl_skbs(struct s3_conn *c3cn)
+{
+	if (c3cn->cpl_close)
+		kfree_skb(c3cn->cpl_close);
+	if (c3cn->cpl_abort_req)
+		kfree_skb(c3cn->cpl_abort_req);
+	if (c3cn->cpl_abort_rpl)
+		kfree_skb(c3cn->cpl_abort_rpl);
+}
+
+static int c3cn_alloc_cpl_skbs(struct s3_conn *c3cn)
+{
+	c3cn->cpl_close = alloc_skb(sizeof(struct cpl_close_con_req),
+				   GFP_KERNEL);
+	if (!c3cn->cpl_close)
+		return -ENOMEM;
+	skb_put(c3cn->cpl_close, sizeof(struct cpl_close_con_req));
+
+	c3cn->cpl_abort_req = alloc_skb(sizeof(struct cpl_abort_req),
+					GFP_KERNEL);
+	if (!c3cn->cpl_abort_req)
+		goto free_cpl_skbs;
+	skb_put(c3cn->cpl_abort_req, sizeof(struct cpl_abort_req));
+
+	c3cn->cpl_abort_rpl = alloc_skb(sizeof(struct cpl_abort_rpl),
+					GFP_KERNEL);
+	if (!c3cn->cpl_abort_rpl)
+		goto free_cpl_skbs;
+	skb_put(c3cn->cpl_abort_rpl, sizeof(struct cpl_abort_rpl));
+
+	return 0;
+
+free_cpl_skbs:
+	c3cn_free_cpl_skbs(c3cn);
+	return -ENOMEM;
+}
+
+/**
+ * c3cn_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 c3cn_release_offload_resources(struct s3_conn *c3cn)
+{
+	struct t3cdev *cdev = c3cn->cdev;
+	unsigned int tid = c3cn->tid;
+
+	if (!cdev)
+		return;
+
+	c3cn->qset = 0;
+
+	c3cn_free_cpl_skbs(c3cn);
+
+	if (c3cn->wr_avail != c3cn->wr_max) {
+		purge_wr_queue(c3cn);
+		reset_wr_list(c3cn);
+	}
+
+	if (c3cn->l2t) {
+		l2t_release(L2DATA(cdev), c3cn->l2t);
+		c3cn->l2t = NULL;
+	}
+
+	if (c3cn->state == C3CN_STATE_CONNECTING) /* we have ATID */
+		s3_free_atid(cdev, tid);
+	else {		/* we have TID */
+		cxgb3_remove_tid(cdev, (void *)c3cn, tid);
+		c3cn_put(c3cn);
+	}
+
+	c3cn->cdev = NULL;
+}
+
+/**
+ * cxgb3i_c3cn_create - allocate and initialize an s3_conn structure
+ * returns the s3_conn structure allocated.
+ */
+struct s3_conn *cxgb3i_c3cn_create(void)
+{
+	struct s3_conn *c3cn;
+
+	c3cn = kzalloc(sizeof(*c3cn), GFP_KERNEL);
+	if (!c3cn)
+		return NULL;
+
+	/* pre-allocate close/abort cpl, so we don't need to wait for memory
+	   when close/abort is requested. */
+	if (c3cn_alloc_cpl_skbs(c3cn) < 0)
+		goto free_c3cn;
+
+	c3cn_conn_debug("alloc c3cn 0x%p.\n", c3cn);
+
+	c3cn->flags = 0;
+	spin_lock_init(&c3cn->lock);
+	atomic_set(&c3cn->refcnt, 1);
+	skb_queue_head_init(&c3cn->receive_queue);
+	skb_queue_head_init(&c3cn->write_queue);
+	setup_timer(&c3cn->retry_timer, NULL, (unsigned long)c3cn);
+	rwlock_init(&c3cn->callback_lock);
+
+	return c3cn;
+
+free_c3cn:
+	kfree(c3cn);
+	return NULL;
+}
+
+static void c3cn_active_close(struct s3_conn *c3cn)
+{
+	int data_lost;
+	int close_req = 0;
+
+	c3cn_conn_debug("c3cn 0x%p, state %u, flag 0x%lx.\n",
+			 c3cn, c3cn->state, c3cn->flags);
+
+	dst_confirm(c3cn->dst_cache);
+
+	c3cn_hold(c3cn);
+	spin_lock_bh(&c3cn->lock);
+
+	data_lost = skb_queue_len(&c3cn->receive_queue);
+	__skb_queue_purge(&c3cn->receive_queue);
+
+	switch (c3cn->state) {
+	case C3CN_STATE_CLOSED:
+	case C3CN_STATE_ACTIVE_CLOSE:
+	case C3CN_STATE_CLOSE_WAIT_1:
+	case C3CN_STATE_CLOSE_WAIT_2:
+	case C3CN_STATE_ABORTING:
+		/* nothing need to be done */
+		break;
+	case C3CN_STATE_CONNECTING:
+		/* defer until cpl_act_open_rpl or cpl_act_establish */
+		c3cn_set_flag(c3cn, C3CN_ACTIVE_CLOSE_NEEDED);
+		break;
+	case C3CN_STATE_ESTABLISHED:
+		close_req = 1;
+		c3cn_set_state(c3cn, C3CN_STATE_ACTIVE_CLOSE);
+		break;
+	case C3CN_STATE_PASSIVE_CLOSE:
+		close_req = 1;
+		c3cn_set_state(c3cn, C3CN_STATE_CLOSE_WAIT_2);
+		break;
+	}
+
+	if (close_req) {
+		if (data_lost)
+			/* Unread data was tossed, zap the connection. */
+			send_abort_req(c3cn);
+		else
+			send_close_req(c3cn);
+	}
+
+	spin_unlock_bh(&c3cn->lock);
+	c3cn_put(c3cn);
+}
+
+/**
+ * cxgb3i_c3cn_release - close and release an iscsi tcp connection and any
+ * 	resource held
+ * @c3cn: the iscsi tcp connection
+ */
+void cxgb3i_c3cn_release(struct s3_conn *c3cn)
+{
+	c3cn_conn_debug("c3cn 0x%p, s %u, f 0x%lx.\n",
+			c3cn, c3cn->state, c3cn->flags);
+	if (likely(c3cn->state != C3CN_STATE_CONNECTING))
+		c3cn_active_close(c3cn);
+	else
+		c3cn_set_flag(c3cn, C3CN_ACTIVE_CLOSE_NEEDED);
+	c3cn_put(c3cn);
+}
+
+static int is_cxgb3_dev(struct net_device *dev)
+{
+	struct cxgb3i_sdev_data *cdata;
+
+	write_lock(&cdata_rwlock);
+	list_for_each_entry(cdata, &cdata_list, list) {
+		struct adap_ports *ports = &cdata->ports;
+		int i;
+
+		for (i = 0; i < ports->nports; i++)
+			if (dev == ports->lldevs[i]) {
+				write_unlock(&cdata_rwlock);
+				return 1;
+			}
+	}
+	write_unlock(&cdata_rwlock);
+	return 0;
+}
+
+/**
+ * cxgb3_egress_dev - return the cxgb3 egress device
+ * @root_dev: the root device anchoring the search
+ * @c3cn: the connection used to determine egress port in bonding mode
+ * @context: in bonding mode, indicates a connection set up or failover
+ *
+ * Return egress device or NULL if the egress device isn't one of our ports.
+ */
+static struct net_device *cxgb3_egress_dev(struct net_device *root_dev,
+					   struct s3_conn *c3cn,
+					   int context)
+{
+	while (root_dev) {
+		if (root_dev->priv_flags & IFF_802_1Q_VLAN)
+			root_dev = vlan_dev_real_dev(root_dev);
+		else if (is_cxgb3_dev(root_dev))
+			return root_dev;
+		else
+			return NULL;
+	}
+	return NULL;
+}
+
+static struct rtable *find_route(__be32 saddr, __be32 daddr,
+				 __be16 sport, __be16 dport)
+{
+	struct rtable *rt;
+	struct flowi fl = {
+		.oif = 0,
+		.nl_u = {
+			 .ip4_u = {
+				   .daddr = daddr,
+				   .saddr = saddr,
+				   .tos = 0 } },
+		.proto = IPPROTO_TCP,
+		.uli_u = {
+			  .ports = {
+				    .sport = sport,
+				    .dport = dport } } };
+
+	if (ip_route_output_flow(&init_net, &rt, &fl, NULL, 0))
+		return NULL;
+	return rt;
+}
+
+/*
+ * Assign offload parameters to some connection fields.
+ */
+static void init_offload_conn(struct s3_conn *c3cn,
+			      struct t3cdev *cdev,
+			      struct dst_entry *dst)
+{
+	BUG_ON(c3cn->cdev != cdev);
+	c3cn->wr_max = c3cn->wr_avail = T3C_DATA(cdev)->max_wrs;
+	c3cn->wr_unacked = 0;
+	c3cn->mss_idx = select_mss(c3cn, dst_mtu(dst));
+
+	reset_wr_list(c3cn);
+}
+
+static int initiate_act_open(struct s3_conn *c3cn, struct net_device *dev)
+{
+	struct cxgb3i_sdev_data *cdata = NDEV2CDATA(dev);
+	struct t3cdev *cdev = cdata->cdev;
+	struct dst_entry *dst = c3cn->dst_cache;
+	struct sk_buff *skb;
+
+	c3cn_conn_debug("c3cn 0x%p, state %u, flag 0x%lx.\n",
+			c3cn, c3cn->state, c3cn->flags);
+	/*
+	 * Initialize connection data.  Note that the flags and ULP mode are
+	 * initialized higher up ...
+	 */
+	c3cn->dev = dev;
+	c3cn->cdev = cdev;
+	c3cn->tid = cxgb3_alloc_atid(cdev, cdata->client, c3cn);
+	if (c3cn->tid < 0)
+		goto out_err;
+
+	c3cn->qset = 0;
+	c3cn->l2t = t3_l2t_get(cdev, dst->neighbour, dev);
+	if (!c3cn->l2t)
+		goto free_tid;
+
+	skb = alloc_skb(sizeof(struct cpl_act_open_req), GFP_KERNEL);
+	if (!skb)
+		goto free_l2t;
+
+	skb->sk = (struct sock *)c3cn;
+	set_arp_failure_handler(skb, act_open_req_arp_failure);
+
+	c3cn_hold(c3cn);
+
+	init_offload_conn(c3cn, cdev, dst);
+	c3cn->err = 0;
+
+	make_act_open_req(c3cn, skb, c3cn->tid, c3cn->l2t);
+	l2t_send(cdev, skb, c3cn->l2t);
+	return 0;
+
+free_l2t:
+	l2t_release(L2DATA(cdev), c3cn->l2t);
+free_tid:
+	s3_free_atid(cdev, c3cn->tid);
+	c3cn->tid = 0;
+out_err:
+	return -1;
+}
+
+
+/**
+ * cxgb3i_c3cn_connect - initiates an iscsi tcp connection to a given address
+ * @c3cn: the iscsi tcp connection
+ * @usin: destination address
+ *
+ * return 0 if active open request is sent, < 0 otherwise.
+ */
+int cxgb3i_c3cn_connect(struct s3_conn *c3cn, struct sockaddr_in *usin)
+{
+	struct rtable *rt;
+	struct net_device *dev;
+	struct cxgb3i_sdev_data *cdata;
+	struct t3cdev *cdev;
+	__be32 sipv4;
+	int err;
+
+	if (usin->sin_family != AF_INET)
+		return -EAFNOSUPPORT;
+
+	c3cn->daddr.sin_port = usin->sin_port;
+	c3cn->daddr.sin_addr.s_addr = usin->sin_addr.s_addr;
+
+	rt = find_route(c3cn->saddr.sin_addr.s_addr,
+			c3cn->daddr.sin_addr.s_addr,
+			c3cn->saddr.sin_port,
+			c3cn->daddr.sin_port);
+	if (rt == NULL) {
+		c3cn_conn_debug("NO route to 0x%x, port %u.\n",
+				c3cn->daddr.sin_addr.s_addr,
+				ntohs(c3cn->daddr.sin_port));
+		return -ENETUNREACH;
+	}
+
+	if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
+		c3cn_conn_debug("multi-cast route to 0x%x, port %u.\n",
+				c3cn->daddr.sin_addr.s_addr,
+				ntohs(c3cn->daddr.sin_port));
+		ip_rt_put(rt);
+		return -ENETUNREACH;
+	}
+
+	if (!c3cn->saddr.sin_addr.s_addr)
+		c3cn->saddr.sin_addr.s_addr = rt->rt_src;
+
+	/* now commit destination to connection */
+	c3cn->dst_cache = &rt->u.dst;
+
+	/* try to establish an offloaded connection */
+	dev = cxgb3_egress_dev(c3cn->dst_cache->dev, c3cn, 0);
+	if (dev == NULL) {
+		c3cn_conn_debug("c3cn 0x%p, egress dev NULL.\n", c3cn);
+		return -ENETUNREACH;
+	}
+	cdata = NDEV2CDATA(dev);
+	cdev = cdata->cdev;
+
+	/* get a source port if one hasn't been provided */
+	err = c3cn_get_port(c3cn, cdata);
+	if (err)
+		return err;
+
+	c3cn_conn_debug("c3cn 0x%p get port %u.\n",
+			c3cn, ntohs(c3cn->saddr.sin_port));
+
+	sipv4 = cxgb3i_get_private_ipv4addr(dev);
+	if (!sipv4) {
+		c3cn_conn_debug("c3cn 0x%p, iscsi ip not configured.\n", c3cn);
+		sipv4 = c3cn->saddr.sin_addr.s_addr;
+		cxgb3i_set_private_ipv4addr(dev, sipv4);
+	} else
+		c3cn->saddr.sin_addr.s_addr = sipv4;
+
+	c3cn_conn_debug("c3cn 0x%p, %u.%u.%u.%u,%u-%u.%u.%u.%u,%u SYN_SENT.\n",
+			c3cn, NIPQUAD(c3cn->saddr.sin_addr.s_addr),
+			ntohs(c3cn->saddr.sin_port),
+			NIPQUAD(c3cn->daddr.sin_addr.s_addr),
+			ntohs(c3cn->daddr.sin_port));
+
+	c3cn_set_state(c3cn, C3CN_STATE_CONNECTING);
+	if (!initiate_act_open(c3cn, dev))
+		return 0;
+
+	/*
+	 * If we get here, we don't have an offload connection so simply
+	 * return a failure.
+	 */
+	err = -ENOTSUPP;
+
+	/*
+	 * This trashes the connection and releases the local port,
+	 * if necessary.
+	 */
+	c3cn_conn_debug("c3cn 0x%p -> CLOSED.\n", c3cn);
+	c3cn_set_state(c3cn, C3CN_STATE_CLOSED);
+	ip_rt_put(rt);
+	c3cn_put_port(c3cn);
+	c3cn->daddr.sin_port = 0;
+	return err;
+}
+
+/**
+ * cxgb3i_c3cn_rx_credits - ack received tcp data.
+ * @c3cn: iscsi tcp connection
+ * @copied: # of bytes processed
+ *
+ * Called after some received data has been read.  It returns RX credits
+ * to the HW for the amount of data processed.
+ */
+void cxgb3i_c3cn_rx_credits(struct s3_conn *c3cn, int copied)
+{
+	struct t3cdev *cdev;
+	int must_send;
+	u32 credits, dack = 0;
+
+	if (c3cn->state != C3CN_STATE_ESTABLISHED)
+		return;
+
+	credits = c3cn->copied_seq - c3cn->rcv_wup;
+	if (unlikely(!credits))
+		return;
+
+	cdev = c3cn->cdev;
+
+	if (unlikely(cxgb3_rx_credit_thres == 0))
+		return;
+
+	dack = F_RX_DACK_CHANGE | V_RX_DACK_MODE(1);
+
+	/*
+	 * For coalescing to work effectively ensure the receive window has
+	 * at least 16KB left.
+	 */
+	must_send = credits + 16384 >= cxgb3_rcv_win;
+
+	if (must_send || credits >= cxgb3_rx_credit_thres)
+		c3cn->rcv_wup += send_rx_credits(c3cn, credits, dack);
+}
+
+/**
+ * cxgb3i_c3cn_send_pdus - send the skbs containing iscsi pdus
+ * @c3cn: iscsi tcp connection
+ * @skb: skb contains the iscsi pdu
+ *
+ * Add a list of skbs to a connection send queue. The skbs must comply with
+ * the max size limit of the device and have a headroom of at least
+ * TX_HEADER_LEN bytes.
+ * Return # of bytes queued.
+ */
+int cxgb3i_c3cn_send_pdus(struct s3_conn *c3cn, struct sk_buff *skb)
+{
+	struct sk_buff *next;
+	int err, copied = 0;
+
+	spin_lock_bh(&c3cn->lock);
+
+	if (c3cn->state != C3CN_STATE_ESTABLISHED) {
+		c3cn_tx_debug("c3cn 0x%p, not in est. state %u.\n",
+			      c3cn, c3cn->state);
+		err = -EAGAIN;
+		goto out_err;
+	}
+
+	err = -EPIPE;
+	if (c3cn->err) {
+		c3cn_tx_debug("c3cn 0x%p, err %d.\n", c3cn, c3cn->err);
+		goto out_err;
+	}
+
+	while (skb) {
+		int frags = skb_shinfo(skb)->nr_frags +
+				(skb->len != skb->data_len);
+
+		if (unlikely(skb_headroom(skb) < TX_HEADER_LEN)) {
+			c3cn_tx_debug("c3cn 0x%p, skb head.\n", c3cn);
+			err = -EINVAL;
+			goto out_err;
+		}
+
+		if (frags >= SKB_WR_LIST_SIZE) {
+			cxgb3i_log_error("c3cn 0x%p, tx frags %d, len %u,%u.\n",
+					 c3cn, skb_shinfo(skb)->nr_frags,
+					 skb->len, skb->data_len);
+			err = -EINVAL;
+			goto out_err;
+		}
+
+		next = skb->next;
+		skb->next = NULL;
+		skb_entail(c3cn, skb, C3CB_FLAG_NO_APPEND | C3CB_FLAG_NEED_HDR);
+		copied += skb->len;
+		c3cn->write_seq += skb->len + ulp_extra_len(skb);
+		skb = next;
+	}
+done:
+	if (likely(skb_queue_len(&c3cn->write_queue)))
+		c3cn_push_tx_frames(c3cn, 1);
+	spin_unlock_bh(&c3cn->lock);
+	return copied;
+
+out_err:
+	if (copied == 0 && err == -EPIPE)
+		copied = c3cn->err ? c3cn->err : -EPIPE;
+	goto done;
+}
+
+static void sdev_data_cleanup(struct cxgb3i_sdev_data *cdata)
+{
+	struct adap_ports *ports = &cdata->ports;
+	int i;
+
+	for (i = 0; i < ports->nports; i++)
+		NDEV2CDATA(ports->lldevs[i]) = NULL;
+	cxgb3i_free_big_mem(cdata);
+}
+
+void cxgb3i_sdev_cleanup(void)
+{
+	struct cxgb3i_sdev_data *cdata;
+
+	write_lock(&cdata_rwlock);
+	list_for_each_entry(cdata, &cdata_list, list) {
+		list_del(&cdata->list);
+		sdev_data_cleanup(cdata);
+	}
+	write_unlock(&cdata_rwlock);
+}
+
+int cxgb3i_sdev_init(cxgb3_cpl_handler_func *cpl_handlers)
+{
+	cpl_handlers[CPL_ACT_ESTABLISH] = do_act_establish;
+	cpl_handlers[CPL_ACT_OPEN_RPL] = do_act_open_rpl;
+	cpl_handlers[CPL_PEER_CLOSE] = do_peer_close;
+	cpl_handlers[CPL_ABORT_REQ_RSS] = do_abort_req;
+	cpl_handlers[CPL_ABORT_RPL_RSS] = do_abort_rpl;
+	cpl_handlers[CPL_CLOSE_CON_RPL] = do_close_con_rpl;
+	cpl_handlers[CPL_TX_DMA_ACK] = do_wr_ack;
+	cpl_handlers[CPL_ISCSI_HDR] = do_iscsi_hdr;
+
+	if (cxgb3_max_connect > CXGB3I_MAX_CONN)
+		cxgb3_max_connect = CXGB3I_MAX_CONN;
+	return 0;
+}
+
+/**
+ * cxgb3i_sdev_add - allocate and initialize resources for each adapter found
+ * @cdev:	t3cdev adapter
+ * @client:	cxgb3 driver client
+ */
+void cxgb3i_sdev_add(struct t3cdev *cdev, struct cxgb3_client *client)
+{
+	struct cxgb3i_sdev_data *cdata;
+	struct ofld_page_info rx_page_info;
+	unsigned int wr_len;
+	int mapsize = DIV_ROUND_UP(cxgb3_max_connect,
+				   8 * sizeof(unsigned long));
+	int i;
+
+	cdata =  cxgb3i_alloc_big_mem(sizeof(*cdata) + mapsize, GFP_KERNEL);
+	if (!cdata)
+		return;
+
+	if (cdev->ctl(cdev, GET_WR_LEN, &wr_len) < 0 ||
+	    cdev->ctl(cdev, GET_PORTS, &cdata->ports) < 0 ||
+	    cdev->ctl(cdev, GET_RX_PAGE_INFO, &rx_page_info) < 0)
+		goto free_cdata;
+
+	s3_init_wr_tab(wr_len);
+
+	INIT_LIST_HEAD(&cdata->list);
+	cdata->cdev = cdev;
+	cdata->client = client;
+
+	for (i = 0; i < cdata->ports.nports; i++)
+		NDEV2CDATA(cdata->ports.lldevs[i]) = cdata;
+
+	write_lock(&cdata_rwlock);
+	list_add_tail(&cdata->list, &cdata_list);
+	write_unlock(&cdata_rwlock);
+
+	return;
+
+free_cdata:
+	cxgb3i_free_big_mem(cdata);
+}
+
+/**
+ * cxgb3i_sdev_remove - free the allocated resources for the adapter
+ * @cdev:	t3cdev adapter
+ */
+void cxgb3i_sdev_remove(struct t3cdev *cdev)
+{
+	struct cxgb3i_sdev_data *cdata = CXGB3_SDEV_DATA(cdev);
+
+	write_lock(&cdata_rwlock);
+	list_del(&cdata->list);
+	write_unlock(&cdata_rwlock);
+
+	sdev_data_cleanup(cdata);
+}