Merge branch 'cxgb4'

Hariprasad Shenai says:

====================
Adds support for CIQ and other misc. fixes for rdma/cxgb4

This patch series adds support to allocate and use IQs specifically for
indirect interrupts, adds fixes to align ISS for iWARP connections & fixes
related to tcp snd/rvd window for Chelsio T4/T5 adapters on iw_cxgb4.
Also changes Interrupt Holdoff Packet Count threshold of response queues for
cxgb4 driver.

The patches series is created against 'net-next' tree.
And includes patches on cxgb4 and iw_cxgb4 driver.

Since this patch-series contains cxgb4 and iw_cxgb4 patches, we would like to
request this patch series to get merged via David Miller's 'net-next' tree.

We have included all the maintainers of respective drivers. Kindly review the
change and let us know in case of any review comments.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

12 files changed, 356 insertions, 57 deletions

diff --git a/drivers/infiniband/hw/cxgb4/cm.c b/drivers/infiniband/hw/cxgb4/cm.c
index 1f863a96a48..965eaafd585 100644
--- a/drivers/infiniband/hw/cxgb4/cm.c
+++ b/drivers/infiniband/hw/cxgb4/cm.c
@@ -232,12 +232,16 @@ static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
 
 static void set_emss(struct c4iw_ep *ep, u16 opt)
 {
-	ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] - 40;
+	ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] -
+		   sizeof(struct iphdr) - sizeof(struct tcphdr);
 	ep->mss = ep->emss;
 	if (GET_TCPOPT_TSTAMP(opt))
 		ep->emss -= 12;
 	if (ep->emss < 128)
 		ep->emss = 128;
+	if (ep->emss & 7)
+		PDBG("Warning: misaligned mtu idx %u mss %u emss=%u\n",
+		     GET_TCPOPT_MSS(opt), ep->mss, ep->emss);
 	PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, GET_TCPOPT_MSS(opt),
 	     ep->mss, ep->emss);
 }
@@ -468,7 +472,7 @@ static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
 	flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
 	flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
 	flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
-	flowc->mnemval[6].val = cpu_to_be32(snd_win);
+	flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
 	flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
 	flowc->mnemval[7].val = cpu_to_be32(ep->emss);
 	/* Pad WR to 16 byte boundary */
@@ -528,6 +532,17 @@ static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
 }
 
+static void best_mtu(const unsigned short *mtus, unsigned short mtu,
+		     unsigned int *idx, int use_ts)
+{
+	unsigned short hdr_size = sizeof(struct iphdr) +
+				  sizeof(struct tcphdr) +
+				  (use_ts ? 12 : 0);
+	unsigned short data_size = mtu - hdr_size;
+
+	cxgb4_best_aligned_mtu(mtus, hdr_size, data_size, 8, idx);
+}
+
 static int send_connect(struct c4iw_ep *ep)
 {
 	struct cpl_act_open_req *req;
@@ -550,6 +565,7 @@ static int send_connect(struct c4iw_ep *ep)
 	struct sockaddr_in *ra = (struct sockaddr_in *)&ep->com.remote_addr;
 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
+	int win;
 
 	wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
 			roundup(sizev4, 16) :
@@ -565,8 +581,18 @@ static int send_connect(struct c4iw_ep *ep)
 	}
 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
 
-	cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
+	best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
+		 enable_tcp_timestamps);
 	wscale = compute_wscale(rcv_win);
+
+	/*
+	 * Specify the largest window that will fit in opt0. The
+	 * remainder will be specified in the rx_data_ack.
+	 */
+	win = ep->rcv_win >> 10;
+	if (win > RCV_BUFSIZ_MASK)
+		win = RCV_BUFSIZ_MASK;
+
 	opt0 = (nocong ? NO_CONG(1) : 0) |
 	       KEEP_ALIVE(1) |
 	       DELACK(1) |
@@ -577,7 +603,7 @@ static int send_connect(struct c4iw_ep *ep)
 	       SMAC_SEL(ep->smac_idx) |
 	       DSCP(ep->tos) |
 	       ULP_MODE(ULP_MODE_TCPDDP) |
-	       RCV_BUFSIZ(rcv_win>>10);
+	       RCV_BUFSIZ(win);
 	opt2 = RX_CHANNEL(0) |
 	       CCTRL_ECN(enable_ecn) |
 	       RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
@@ -633,6 +659,13 @@ static int send_connect(struct c4iw_ep *ep)
 			req6->opt2 = cpu_to_be32(opt2);
 		}
 	} else {
+		u32 isn = (prandom_u32() & ~7UL) - 1;
+
+		opt2 |= T5_OPT_2_VALID;
+		opt2 |= CONG_CNTRL_VALID; /* OPT_2_ISS for T5 */
+		if (peer2peer)
+			isn += 4;
+
 		if (ep->com.remote_addr.ss_family == AF_INET) {
 			t5_req = (struct cpl_t5_act_open_req *)
 				 skb_put(skb, wrlen);
@@ -649,6 +682,9 @@ static int send_connect(struct c4iw_ep *ep)
 						     cxgb4_select_ntuple(
 					     ep->com.dev->rdev.lldi.ports[0],
 					     ep->l2t)));
+			t5_req->rsvd = cpu_to_be32(isn);
+			PDBG("%s snd_isn %u\n", __func__,
+			     be32_to_cpu(t5_req->rsvd));
 			t5_req->opt2 = cpu_to_be32(opt2);
 		} else {
 			t5_req6 = (struct cpl_t5_act_open_req6 *)
@@ -672,6 +708,9 @@ static int send_connect(struct c4iw_ep *ep)
 							cxgb4_select_ntuple(
 						ep->com.dev->rdev.lldi.ports[0],
 						ep->l2t));
+			t5_req6->rsvd = cpu_to_be32(isn);
+			PDBG("%s snd_isn %u\n", __func__,
+			     be32_to_cpu(t5_req6->rsvd));
 			t5_req6->opt2 = cpu_to_be32(opt2);
 		}
 	}
@@ -1145,6 +1184,14 @@ static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
 		return 0;
 	}
 
+	/*
+	 * If we couldn't specify the entire rcv window at connection setup
+	 * due to the limit in the number of bits in the RCV_BUFSIZ field,
+	 * then add the overage in to the credits returned.
+	 */
+	if (ep->rcv_win > RCV_BUFSIZ_MASK * 1024)
+		credits += ep->rcv_win - RCV_BUFSIZ_MASK * 1024;
+
 	req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
 	memset(req, 0, wrlen);
 	INIT_TP_WR(req, ep->hwtid);
@@ -1618,6 +1665,7 @@ static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
 	unsigned int mtu_idx;
 	int wscale;
 	struct sockaddr_in *sin;
+	int win;
 
 	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
 	req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
@@ -1640,8 +1688,18 @@ static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
 			htons(F_FW_OFLD_CONNECTION_WR_CPLRXDATAACK);
 	req->tcb.tx_max = (__force __be32) jiffies;
 	req->tcb.rcv_adv = htons(1);
-	cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
+	best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
+		 enable_tcp_timestamps);
 	wscale = compute_wscale(rcv_win);
+
+	/*
+	 * Specify the largest window that will fit in opt0. The
+	 * remainder will be specified in the rx_data_ack.
+	 */
+	win = ep->rcv_win >> 10;
+	if (win > RCV_BUFSIZ_MASK)
+		win = RCV_BUFSIZ_MASK;
+
 	req->tcb.opt0 = (__force __be64) (TCAM_BYPASS(1) |
 		(nocong ? NO_CONG(1) : 0) |
 		KEEP_ALIVE(1) |
@@ -1653,7 +1711,7 @@ static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
 		SMAC_SEL(ep->smac_idx) |
 		DSCP(ep->tos) |
 		ULP_MODE(ULP_MODE_TCPDDP) |
-		RCV_BUFSIZ(rcv_win >> 10));
+		RCV_BUFSIZ(win));
 	req->tcb.opt2 = (__force __be32) (PACE(1) |
 		TX_QUEUE(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
 		RX_CHANNEL(0) |
@@ -1690,6 +1748,13 @@ static int is_neg_adv(unsigned int status)
 	       status == CPL_ERR_KEEPALV_NEG_ADVICE;
 }
 
+static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
+{
+	ep->snd_win = snd_win;
+	ep->rcv_win = rcv_win;
+	PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win);
+}
+
 #define ACT_OPEN_RETRY_COUNT 2
 
 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
@@ -1738,6 +1803,7 @@ static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
 		ep->ctrlq_idx = cxgb4_port_idx(pdev);
 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
 			cxgb4_port_idx(pdev) * step];
+		set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
 		dev_put(pdev);
 	} else {
 		pdev = get_real_dev(n->dev);
@@ -1756,6 +1822,7 @@ static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
 			cdev->rdev.lldi.nchan;
 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
 			cxgb4_port_idx(n->dev) * step];
+		set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
 
 		if (clear_mpa_v1) {
 			ep->retry_with_mpa_v1 = 0;
@@ -1986,13 +2053,36 @@ static void accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
 	u64 opt0;
 	u32 opt2;
 	int wscale;
+	struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
+	int win;
 
 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
 	BUG_ON(skb_cloned(skb));
-	skb_trim(skb, sizeof(*rpl));
+
 	skb_get(skb);
-	cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
+	rpl = cplhdr(skb);
+	if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
+		skb_trim(skb, roundup(sizeof(*rpl5), 16));
+		rpl5 = (void *)rpl;
+		INIT_TP_WR(rpl5, ep->hwtid);
+	} else {
+		skb_trim(skb, sizeof(*rpl));
+		INIT_TP_WR(rpl, ep->hwtid);
+	}
+	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
+						    ep->hwtid));
+
+	best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
+		 enable_tcp_timestamps && req->tcpopt.tstamp);
 	wscale = compute_wscale(rcv_win);
+
+	/*
+	 * Specify the largest window that will fit in opt0. The
+	 * remainder will be specified in the rx_data_ack.
+	 */
+	win = ep->rcv_win >> 10;
+	if (win > RCV_BUFSIZ_MASK)
+		win = RCV_BUFSIZ_MASK;
 	opt0 = (nocong ? NO_CONG(1) : 0) |
 	       KEEP_ALIVE(1) |
 	       DELACK(1) |
@@ -2003,7 +2093,7 @@ static void accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
 	       SMAC_SEL(ep->smac_idx) |
 	       DSCP(ep->tos >> 2) |
 	       ULP_MODE(ULP_MODE_TCPDDP) |
-	       RCV_BUFSIZ(rcv_win>>10);
+	       RCV_BUFSIZ(win);
 	opt2 = RX_CHANNEL(0) |
 	       RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
 
@@ -2023,14 +2113,18 @@ static void accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
 			opt2 |= CCTRL_ECN(1);
 	}
 	if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
+		u32 isn = (prandom_u32() & ~7UL) - 1;
 		opt2 |= T5_OPT_2_VALID;
 		opt2 |= V_CONG_CNTRL(CONG_ALG_TAHOE);
+		opt2 |= CONG_CNTRL_VALID; /* OPT_2_ISS for T5 */
+		rpl5 = (void *)rpl;
+		memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
+		if (peer2peer)
+			isn += 4;
+		rpl5->iss = cpu_to_be32(isn);
+		PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
 	}
 
-	rpl = cplhdr(skb);
-	INIT_TP_WR(rpl, ep->hwtid);
-	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
-				      ep->hwtid));
 	rpl->opt0 = cpu_to_be64(opt0);
 	rpl->opt2 = cpu_to_be32(opt2);
 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
@@ -2095,6 +2189,7 @@ static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
 	int err;
 	u16 peer_mss = ntohs(req->tcpopt.mss);
 	int iptype;
+	unsigned short hdrs;
 
 	parent_ep = lookup_stid(t, stid);
 	if (!parent_ep) {
@@ -2152,8 +2247,10 @@ static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
 		goto reject;
 	}
 
-	if (peer_mss && child_ep->mtu > (peer_mss + 40))
-		child_ep->mtu = peer_mss + 40;
+	hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) +
+	       ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
+	if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
+		child_ep->mtu = peer_mss + hdrs;
 
 	state_set(&child_ep->com, CONNECTING);
 	child_ep->com.dev = dev;
diff --git a/drivers/infiniband/hw/cxgb4/cq.c b/drivers/infiniband/hw/cxgb4/cq.c
index cfaa56ada18..71fc2ef203f 100644
--- a/drivers/infiniband/hw/cxgb4/cq.c
+++ b/drivers/infiniband/hw/cxgb4/cq.c
@@ -134,7 +134,8 @@ static int create_cq(struct c4iw_rdev *rdev, struct t4_cq *cq,
 			V_FW_RI_RES_WR_IQANUS(0) |
 			V_FW_RI_RES_WR_IQANUD(1) |
 			F_FW_RI_RES_WR_IQANDST |
-			V_FW_RI_RES_WR_IQANDSTINDEX(*rdev->lldi.rxq_ids));
+			V_FW_RI_RES_WR_IQANDSTINDEX(
+				rdev->lldi.ciq_ids[cq->vector]));
 	res->u.cq.iqdroprss_to_iqesize = cpu_to_be16(
 			F_FW_RI_RES_WR_IQDROPRSS |
 			V_FW_RI_RES_WR_IQPCIECH(2) |
@@ -870,6 +871,9 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev, int entries,
 
 	rhp = to_c4iw_dev(ibdev);
 
+	if (vector >= rhp->rdev.lldi.nciq)
+		return ERR_PTR(-EINVAL);
+
 	chp = kzalloc(sizeof(*chp), GFP_KERNEL);
 	if (!chp)
 		return ERR_PTR(-ENOMEM);
@@ -915,6 +919,7 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev, int entries,
 	}
 	chp->cq.size = hwentries;
 	chp->cq.memsize = memsize;
+	chp->cq.vector = vector;
 
 	ret = create_cq(&rhp->rdev, &chp->cq,
 			ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
diff --git a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
index 7474b490760..7493dfeb812 100644
--- a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
+++ b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
@@ -805,6 +805,8 @@ struct c4iw_ep {
 	u8 retry_with_mpa_v1;
 	u8 tried_with_mpa_v1;
 	unsigned int retry_count;
+	int snd_win;
+	int rcv_win;
 };
 
 static inline struct c4iw_ep *to_ep(struct iw_cm_id *cm_id)
diff --git a/drivers/infiniband/hw/cxgb4/provider.c b/drivers/infiniband/hw/cxgb4/provider.c
index a94a3e12c34..31cd1882c47 100644
--- a/drivers/infiniband/hw/cxgb4/provider.c
+++ b/drivers/infiniband/hw/cxgb4/provider.c
@@ -499,7 +499,7 @@ int c4iw_register_device(struct c4iw_dev *dev)
 	dev->ibdev.node_type = RDMA_NODE_RNIC;
 	memcpy(dev->ibdev.node_desc, C4IW_NODE_DESC, sizeof(C4IW_NODE_DESC));
 	dev->ibdev.phys_port_cnt = dev->rdev.lldi.nports;
-	dev->ibdev.num_comp_vectors = 1;
+	dev->ibdev.num_comp_vectors =  dev->rdev.lldi.nciq;
 	dev->ibdev.dma_device = &(dev->rdev.lldi.pdev->dev);
 	dev->ibdev.query_device = c4iw_query_device;
 	dev->ibdev.query_port = c4iw_query_port;
diff --git a/drivers/infiniband/hw/cxgb4/t4.h b/drivers/infiniband/hw/cxgb4/t4.h
index 2178f319841..68b0a6bf4eb 100644
--- a/drivers/infiniband/hw/cxgb4/t4.h
+++ b/drivers/infiniband/hw/cxgb4/t4.h
@@ -542,6 +542,7 @@ struct t4_cq {
 	size_t memsize;
 	__be64 bits_type_ts;
 	u32 cqid;
+	int vector;
 	u16 size; /* including status page */
 	u16 cidx;
 	u16 sw_pidx;
diff --git a/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h b/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h
index 6121ca08fe5..91289a051af 100644
--- a/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h
+++ b/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h
@@ -848,6 +848,7 @@ enum {                     /* TCP congestion control algorithms */
 #define V_CONG_CNTRL(x) ((x) << S_CONG_CNTRL)
 #define G_CONG_CNTRL(x) (((x) >> S_CONG_CNTRL) & M_CONG_CNTRL)
 
+#define CONG_CNTRL_VALID   (1 << 18)
 #define T5_OPT_2_VALID       (1 << 31)
 
 #endif /* _T4FW_RI_API_H_ */
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
index 32db3770926..f503dce4ab1 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
@@ -357,11 +357,17 @@ enum {
 	MAX_OFLD_QSETS = 16,          /* # of offload Tx/Rx queue sets */
 	MAX_CTRL_QUEUES = NCHAN,      /* # of control Tx queues */
 	MAX_RDMA_QUEUES = NCHAN,      /* # of streaming RDMA Rx queues */
+	MAX_RDMA_CIQS = NCHAN,        /* # of  RDMA concentrator IQs */
+	MAX_ISCSI_QUEUES = NCHAN,     /* # of streaming iSCSI Rx queues */
 };
 
 enum {
-	MAX_EGRQ = 128,         /* max # of egress queues, including FLs */
-	MAX_INGQ = 64           /* max # of interrupt-capable ingress queues */
+	INGQ_EXTRAS = 2,        /* firmware event queue and */
+				/*   forwarded interrupts */
+	MAX_EGRQ = MAX_ETH_QSETS*2 + MAX_OFLD_QSETS*2
+		   + MAX_CTRL_QUEUES + MAX_RDMA_QUEUES + MAX_ISCSI_QUEUES,
+	MAX_INGQ = MAX_ETH_QSETS + MAX_OFLD_QSETS + MAX_RDMA_QUEUES
+		   + MAX_RDMA_CIQS + MAX_ISCSI_QUEUES + INGQ_EXTRAS,
 };
 
 struct adapter;
@@ -538,6 +544,7 @@ struct sge {
 	struct sge_eth_rxq ethrxq[MAX_ETH_QSETS];
 	struct sge_ofld_rxq ofldrxq[MAX_OFLD_QSETS];
 	struct sge_ofld_rxq rdmarxq[MAX_RDMA_QUEUES];
+	struct sge_ofld_rxq rdmaciq[MAX_RDMA_CIQS];
 	struct sge_rspq fw_evtq ____cacheline_aligned_in_smp;
 
 	struct sge_rspq intrq ____cacheline_aligned_in_smp;
@@ -548,8 +555,10 @@ struct sge {
 	u16 ethtxq_rover;           /* Tx queue to clean up next */
 	u16 ofldqsets;              /* # of active offload queue sets */
 	u16 rdmaqs;                 /* # of available RDMA Rx queues */
+	u16 rdmaciqs;               /* # of available RDMA concentrator IQs */
 	u16 ofld_rxq[MAX_OFLD_QSETS];
 	u16 rdma_rxq[NCHAN];
+	u16 rdma_ciq[NCHAN];
 	u16 timer_val[SGE_NTIMERS];
 	u8 counter_val[SGE_NCOUNTERS];
 	u32 fl_pg_order;            /* large page allocation size */
@@ -577,6 +586,7 @@ struct sge {
 #define for_each_ethrxq(sge, i) for (i = 0; i < (sge)->ethqsets; i++)
 #define for_each_ofldrxq(sge, i) for (i = 0; i < (sge)->ofldqsets; i++)
 #define for_each_rdmarxq(sge, i) for (i = 0; i < (sge)->rdmaqs; i++)
+#define for_each_rdmaciq(sge, i) for (i = 0; i < (sge)->rdmaciqs; i++)
 
 struct l2t_data;
 
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
index 8cf6be93f49..2f8d6b91038 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
@@ -818,12 +818,17 @@ static void name_msix_vecs(struct adapter *adap)
 	for_each_rdmarxq(&adap->sge, i)
 		snprintf(adap->msix_info[msi_idx++].desc, n, "%s-rdma%d",
 			 adap->port[0]->name, i);
+
+	for_each_rdmaciq(&adap->sge, i)
+		snprintf(adap->msix_info[msi_idx++].desc, n, "%s-rdma-ciq%d",
+			 adap->port[0]->name, i);
 }
 
 static int request_msix_queue_irqs(struct adapter *adap)
 {
 	struct sge *s = &adap->sge;
-	int err, ethqidx, ofldqidx = 0, rdmaqidx = 0, msi_index = 2;
+	int err, ethqidx, ofldqidx = 0, rdmaqidx = 0, rdmaciqqidx = 0;
+	int msi_index = 2;
 
 	err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
 			  adap->msix_info[1].desc, &s->fw_evtq);
@@ -857,9 +862,21 @@ static int request_msix_queue_irqs(struct adapter *adap)
 			goto unwind;
 		msi_index++;
 	}
+	for_each_rdmaciq(s, rdmaciqqidx) {
+		err = request_irq(adap->msix_info[msi_index].vec,
+				  t4_sge_intr_msix, 0,
+				  adap->msix_info[msi_index].desc,
+				  &s->rdmaciq[rdmaciqqidx].rspq);
+		if (err)
+			goto unwind;
+		msi_index++;
+	}
 	return 0;
 
 unwind:
+	while (--rdmaciqqidx >= 0)
+		free_irq(adap->msix_info[--msi_index].vec,
+			 &s->rdmaciq[rdmaciqqidx].rspq);
 	while (--rdmaqidx >= 0)
 		free_irq(adap->msix_info[--msi_index].vec,
 			 &s->rdmarxq[rdmaqidx].rspq);
@@ -885,6 +902,8 @@ static void free_msix_queue_irqs(struct adapter *adap)
 		free_irq(adap->msix_info[msi_index++].vec, &s->ofldrxq[i].rspq);
 	for_each_rdmarxq(s, i)
 		free_irq(adap->msix_info[msi_index++].vec, &s->rdmarxq[i].rspq);
+	for_each_rdmaciq(s, i)
+		free_irq(adap->msix_info[msi_index++].vec, &s->rdmaciq[i].rspq);
 }
 
 /**
@@ -1047,7 +1066,8 @@ freeout:	t4_free_sge_resources(adap);
 		if (msi_idx > 0)
 			msi_idx++;
 		err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev, msi_idx,
-				       &q->fl, uldrx_handler);
+				       q->fl.size ? &q->fl : NULL,
+				       uldrx_handler);
 		if (err)
 			goto freeout;
 		memset(&q->stats, 0, sizeof(q->stats));
@@ -1064,13 +1084,28 @@ freeout:	t4_free_sge_resources(adap);
 		if (msi_idx > 0)
 			msi_idx++;
 		err = t4_sge_alloc_rxq(adap, &q->rspq, false, adap->port[i],
-				       msi_idx, &q->fl, uldrx_handler);
+				       msi_idx, q->fl.size ? &q->fl : NULL,
+				       uldrx_handler);
 		if (err)
 			goto freeout;
 		memset(&q->stats, 0, sizeof(q->stats));
 		s->rdma_rxq[i] = q->rspq.abs_id;
 	}
 
+	for_each_rdmaciq(s, i) {
+		struct sge_ofld_rxq *q = &s->rdmaciq[i];
+
+		if (msi_idx > 0)
+			msi_idx++;
+		err = t4_sge_alloc_rxq(adap, &q->rspq, false, adap->port[i],
+				       msi_idx, q->fl.size ? &q->fl : NULL,
+				       uldrx_handler);
+		if (err)
+			goto freeout;
+		memset(&q->stats, 0, sizeof(q->stats));
+		s->rdma_ciq[i] = q->rspq.abs_id;
+	}
+
 	for_each_port(adap, i) {
 		/*
 		 * Note that ->rdmarxq[i].rspq.cntxt_id below is 0 if we don't
@@ -2468,8 +2503,7 @@ static unsigned int qtimer_val(const struct adapter *adap,
 }
 
 /**
- *	set_rxq_intr_params - set a queue's interrupt holdoff parameters
- *	@adap: the adapter
+ *	set_rspq_intr_params - set a queue's interrupt holdoff parameters
  *	@q: the Rx queue
  *	@us: the hold-off time in us, or 0 to disable timer
  *	@cnt: the hold-off packet count, or 0 to disable counter
@@ -2477,9 +2511,11 @@ static unsigned int qtimer_val(const struct adapter *adap,
  *	Sets an Rx queue's interrupt hold-off time and packet count.  At least
  *	one of the two needs to be enabled for the queue to generate interrupts.
  */
-static int set_rxq_intr_params(struct adapter *adap, struct sge_rspq *q,
-			       unsigned int us, unsigned int cnt)
+static int set_rspq_intr_params(struct sge_rspq *q,
+				unsigned int us, unsigned int cnt)
 {
+	struct adapter *adap = q->adap;
+
 	if ((us | cnt) == 0)
 		cnt = 1;
 
@@ -2506,24 +2542,34 @@ static int set_rxq_intr_params(struct adapter *adap, struct sge_rspq *q,
 	return 0;
 }
 
-static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
+/**
+ * set_rx_intr_params - set a net devices's RX interrupt holdoff paramete!
+ * @dev: the network device
+ * @us: the hold-off time in us, or 0 to disable timer
+ * @cnt: the hold-off packet count, or 0 to disable counter
+ *
+ * Set the RX interrupt hold-off parameters for a network device.
+ */
+static int set_rx_intr_params(struct net_device *dev,
+			      unsigned int us, unsigned int cnt)
 {
-	const struct port_info *pi = netdev_priv(dev);
+	int i, err;
+	struct port_info *pi = netdev_priv(dev);
 	struct adapter *adap = pi->adapter;
-	struct sge_rspq *q;
-	int i;
-	int r = 0;
-
-	for (i = pi->first_qset; i < pi->first_qset + pi->nqsets; i++) {
-		q = &adap->sge.ethrxq[i].rspq;
-		r = set_rxq_intr_params(adap, q, c->rx_coalesce_usecs,
-			c->rx_max_coalesced_frames);
-		if (r) {
-			dev_err(&dev->dev, "failed to set coalesce %d\n", r);
-			break;
-		}
+	struct sge_eth_rxq *q = &adap->sge.ethrxq[pi->first_qset];
+
+	for (i = 0; i < pi->nqsets; i++, q++) {
+		err = set_rspq_intr_params(&q->rspq, us, cnt);
+		if (err)
+			return err;
 	}
-	return r;
+	return 0;
+}
+
+static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
+{
+	return set_rx_intr_params(dev, c->rx_coalesce_usecs,
+				  c->rx_max_coalesced_frames);
 }
 
 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
@@ -3393,6 +3439,77 @@ unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
 EXPORT_SYMBOL(cxgb4_best_mtu);
 
 /**
+ *     cxgb4_best_aligned_mtu - find best MTU, [hopefully] data size aligned
+ *     @mtus: the HW MTU table
+ *     @header_size: Header Size
+ *     @data_size_max: maximum Data Segment Size
+ *     @data_size_align: desired Data Segment Size Alignment (2^N)
+ *     @mtu_idxp: HW MTU Table Index return value pointer (possibly NULL)
+ *
+ *     Similar to cxgb4_best_mtu() but instead of searching the Hardware
+ *     MTU Table based solely on a Maximum MTU parameter, we break that
+ *     parameter up into a Header Size and Maximum Data Segment Size, and
+ *     provide a desired Data Segment Size Alignment.  If we find an MTU in
+ *     the Hardware MTU Table which will result in a Data Segment Size with
+ *     the requested alignment _and_ that MTU isn't "too far" from the
+ *     closest MTU, then we'll return that rather than the closest MTU.
+ */
+unsigned int cxgb4_best_aligned_mtu(const unsigned short *mtus,
+				    unsigned short header_size,
+				    unsigned short data_size_max,
+				    unsigned short data_size_align,
+				    unsigned int *mtu_idxp)
+{
+	unsigned short max_mtu = header_size + data_size_max;
+	unsigned short data_size_align_mask = data_size_align - 1;
+	int mtu_idx, aligned_mtu_idx;
+
+	/* Scan the MTU Table till we find an MTU which is larger than our
+	 * Maximum MTU or we reach the end of the table.  Along the way,
+	 * record the last MTU found, if any, which will result in a Data
+	 * Segment Length matching the requested alignment.
+	 */
+	for (mtu_idx = 0, aligned_mtu_idx = -1; mtu_idx < NMTUS; mtu_idx++) {
+		unsigned short data_size = mtus[mtu_idx] - header_size;
+
+		/* If this MTU minus the Header Size would result in a
+		 * Data Segment Size of the desired alignment, remember it.
+		 */
+		if ((data_size & data_size_align_mask) == 0)
+			aligned_mtu_idx = mtu_idx;
+
+		/* If we're not at the end of the Hardware MTU Table and the
+		 * next element is larger than our Maximum MTU, drop out of
+		 * the loop.
+		 */
+		if (mtu_idx+1 < NMTUS && mtus[mtu_idx+1] > max_mtu)
+			break;
+	}
+
+	/* If we fell out of the loop because we ran to the end of the table,
+	 * then we just have to use the last [largest] entry.
+	 */
+	if (mtu_idx == NMTUS)
+		mtu_idx--;
+
+	/* If we found an MTU which resulted in the requested Data Segment
+	 * Length alignment and that's "not far" from the largest MTU which is
+	 * less than or equal to the maximum MTU, then use that.
+	 */
+	if (aligned_mtu_idx >= 0 &&
+	    mtu_idx - aligned_mtu_idx <= 1)
+		mtu_idx = aligned_mtu_idx;
+
+	/* If the caller has passed in an MTU Index pointer, pass the
+	 * MTU Index back.  Return the MTU value.
+	 */
+	if (mtu_idxp)
+		*mtu_idxp = mtu_idx;
+	return mtus[mtu_idx];
+}
+EXPORT_SYMBOL(cxgb4_best_aligned_mtu);
+
+/**
  *	cxgb4_port_chan - get the HW channel of a port
  *	@dev: the net device for the port
  *
@@ -3789,7 +3906,9 @@ static void uld_attach(struct adapter *adap, unsigned int uld)
 	lli.mtus = adap->params.mtus;
 	if (uld == CXGB4_ULD_RDMA) {
 		lli.rxq_ids = adap->sge.rdma_rxq;
+		lli.ciq_ids = adap->sge.rdma_ciq;
 		lli.nrxq = adap->sge.rdmaqs;
+		lli.nciq = adap->sge.rdmaciqs;
 	} else if (uld == CXGB4_ULD_ISCSI) {
 		lli.rxq_ids = adap->sge.ofld_rxq;
 		lli.nrxq = adap->sge.ofldqsets;
@@ -5535,13 +5654,41 @@ static int adap_init0(struct adapter *adap)
 #undef FW_PARAM_PFVF
 #undef FW_PARAM_DEV
 
-	/*
-	 * These are finalized by FW initialization, load their values now.
+	/* The MTU/MSS Table is initialized by now, so load their values.  If
+	 * we're initializing the adapter, then we'll make any modifications
+	 * we want to the MTU/MSS Table and also initialize the congestion
+	 * parameters.
 	 */
 	t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
-	t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
-		     adap->params.b_wnd);
+	if (state != DEV_STATE_INIT) {
+		int i;
+
+		/* The default MTU Table contains values 1492 and 1500.
+		 * However, for TCP, it's better to have two values which are
+		 * a multiple of 8 +/- 4 bytes apart near this popular MTU.
+		 * This allows us to have a TCP Data Payload which is a
+		 * multiple of 8 regardless of what combination of TCP Options
+		 * are in use (always a multiple of 4 bytes) which is
+		 * important for performance reasons.  For instance, if no
+		 * options are in use, then we have a 20-byte IP header and a
+		 * 20-byte TCP header.  In this case, a 1500-byte MSS would
+		 * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
+		 * which is not a multiple of 8.  So using an MSS of 1488 in
+		 * this case results in a TCP Data Payload of 1448 bytes which
+		 * is a multiple of 8.  On the other hand, if 12-byte TCP Time
+		 * Stamps have been negotiated, then an MTU of 1500 bytes
+		 * results in a TCP Data Payload of 1448 bytes which, as
+		 * above, is a multiple of 8 bytes ...
+		 */
+		for (i = 0; i < NMTUS; i++)
+			if (adap->params.mtus[i] == 1492) {
+				adap->params.mtus[i] = 1488;
+				break;
+			}
 
+		t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
+			     adap->params.b_wnd);
+	}
 	t4_init_tp_params(adap);
 	adap->flags |= FW_OK;
 	return 0;
@@ -5676,12 +5823,12 @@ static inline bool is_x_10g_port(const struct link_config *lc)
 	       (lc->supported & FW_PORT_CAP_SPEED_40G) != 0;
 }
 
-static inline void init_rspq(struct sge_rspq *q, u8 timer_idx, u8 pkt_cnt_idx,
+static inline void init_rspq(struct adapter *adap, struct sge_rspq *q,
+			     unsigned int us, unsigned int cnt,
 			     unsigned int size, unsigned int iqe_size)
 {
-	q->intr_params = QINTR_TIMER_IDX(timer_idx) |
-			 (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0);
-	q->pktcnt_idx = pkt_cnt_idx < SGE_NCOUNTERS ? pkt_cnt_idx : 0;
+	q->adap = adap;
+	set_rspq_intr_params(q, us, cnt);
 	q->iqe_len = iqe_size;
 	q->size = size;
 }
@@ -5695,6 +5842,7 @@ static void cfg_queues(struct adapter *adap)
 {
 	struct sge *s = &adap->sge;
 	int i, q10g = 0, n10g = 0, qidx = 0;
+	int ciq_size;
 
 	for_each_port(adap, i)
 		n10g += is_x_10g_port(&adap2pinfo(adap, i)->link_cfg);
@@ -5733,12 +5881,13 @@ static void cfg_queues(struct adapter *adap)
 			s->ofldqsets = adap->params.nports;
 		/* For RDMA one Rx queue per channel suffices */
 		s->rdmaqs = adap->params.nports;
+		s->rdmaciqs = adap->params.nports;
 	}
 
 	for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
 		struct sge_eth_rxq *r = &s->ethrxq[i];
 
-		init_rspq(&r->rspq, 0, 0, 1024, 64);
+		init_rspq(adap, &r->rspq, 5, 10, 1024, 64);
 		r->fl.size = 72;
 	}
 
@@ -5754,7 +5903,7 @@ static void cfg_queues(struct adapter *adap)
 	for (i = 0; i < ARRAY_SIZE(s->ofldrxq); i++) {
 		struct sge_ofld_rxq *r = &s->ofldrxq[i];
 
-		init_rspq(&r->rspq, 0, 0, 1024, 64);
+		init_rspq(adap, &r->rspq, 5, 1, 1024, 64);
 		r->rspq.uld = CXGB4_ULD_ISCSI;
 		r->fl.size = 72;
 	}
@@ -5762,13 +5911,26 @@ static void cfg_queues(struct adapter *adap)
 	for (i = 0; i < ARRAY_SIZE(s->rdmarxq); i++) {
 		struct sge_ofld_rxq *r = &s->rdmarxq[i];
 
-		init_rspq(&r->rspq, 0, 0, 511, 64);
+		init_rspq(adap, &r->rspq, 5, 1, 511, 64);
 		r->rspq.uld = CXGB4_ULD_RDMA;
 		r->fl.size = 72;
 	}
 
-	init_rspq(&s->fw_evtq, 6, 0, 512, 64);
-	init_rspq(&s->intrq, 6, 0, 2 * MAX_INGQ, 64);
+	ciq_size = 64 + adap->vres.cq.size + adap->tids.nftids;
+	if (ciq_size > SGE_MAX_IQ_SIZE) {
+		CH_WARN(adap, "CIQ size too small for available IQs\n");
+		ciq_size = SGE_MAX_IQ_SIZE;
+	}
+
+	for (i = 0; i < ARRAY_SIZE(s->rdmaciq); i++) {
+		struct sge_ofld_rxq *r = &s->rdmaciq[i];
+
+		init_rspq(adap, &r->rspq, 5, 1, ciq_size, 64);
+		r->rspq.uld = CXGB4_ULD_RDMA;
+	}
+
+	init_rspq(adap, &s->fw_evtq, 0, 1, 1024, 64);
+	init_rspq(adap, &s->intrq, 0, 1, 2 * MAX_INGQ, 64);
 }
 
 /*
@@ -5815,9 +5977,9 @@ static int enable_msix(struct adapter *adap)
 
 	want = s->max_ethqsets + EXTRA_VECS;
 	if (is_offload(adap)) {
-		want += s->rdmaqs + s->ofldqsets;
+		want += s->rdmaqs + s->rdmaciqs + s->ofldqsets;
 		/* need nchan for each possible ULD */
-		ofld_need = 2 * nchan;
+		ofld_need = 3 * nchan;
 	}
 	need = adap->params.nports + EXTRA_VECS + ofld_need;
 
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h
index e274a047528..55e9daf7f9d 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h
@@ -232,8 +232,10 @@ struct cxgb4_lld_info {
 	const struct cxgb4_virt_res *vr;     /* assorted HW resources */
 	const unsigned short *mtus;          /* MTU table */
 	const unsigned short *rxq_ids;       /* the ULD's Rx queue ids */
+	const unsigned short *ciq_ids;       /* the ULD's concentrator IQ ids */
 	unsigned short nrxq;                 /* # of Rx queues */
 	unsigned short ntxq;                 /* # of Tx queues */
+	unsigned short nciq;		     /* # of concentrator IQ */
 	unsigned char nchan:4;               /* # of channels */
 	unsigned char nports:4;              /* # of ports */
 	unsigned char wr_cred;               /* WR 16-byte credits */
@@ -274,6 +276,11 @@ unsigned int cxgb4_port_viid(const struct net_device *dev);
 unsigned int cxgb4_port_idx(const struct net_device *dev);
 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
 			    unsigned int *idx);
+unsigned int cxgb4_best_aligned_mtu(const unsigned short *mtus,
+				    unsigned short header_size,
+				    unsigned short data_size_max,
+				    unsigned short data_size_align,
+				    unsigned int *mtu_idxp);
 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
 			 struct tp_tcp_stats *v6);
 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
diff --git a/drivers/net/ethernet/chelsio/cxgb4/sge.c b/drivers/net/ethernet/chelsio/cxgb4/sge.c
index cced1a3d518..58bd213fcaf 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/sge.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/sge.c
@@ -2215,7 +2215,6 @@ int t4_sge_alloc_rxq(struct adapter *adap, struct sge_rspq *iq, bool fwevtq,
 	iq->cntxt_id = ntohs(c.iqid);
 	iq->abs_id = ntohs(c.physiqid);
 	iq->size--;                           /* subtract status entry */
-	iq->adap = adap;
 	iq->netdev = dev;
 	iq->handler = hnd;
 
@@ -2515,6 +2514,10 @@ void t4_free_sge_resources(struct adapter *adap)
 		if (oq->rspq.desc)
 			free_rspq_fl(adap, &oq->rspq, &oq->fl);
 	}
+	for (i = 0, oq = adap->sge.rdmaciq; i < adap->sge.rdmaciqs; i++, oq++) {
+		if (oq->rspq.desc)
+			free_rspq_fl(adap, &oq->rspq, &oq->fl);
+	}
 
 	/* clean up offload Tx queues */
 	for (i = 0; i < ARRAY_SIZE(adap->sge.ofldtxq); i++) {
diff --git a/drivers/net/ethernet/chelsio/cxgb4/t4_hw.h b/drivers/net/ethernet/chelsio/cxgb4/t4_hw.h
index 1d1623be9f1..71b799b5b0f 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/t4_hw.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/t4_hw.h
@@ -68,6 +68,7 @@ enum {
 	SGE_MAX_WR_LEN = 512,     /* max WR size in bytes */
 	SGE_NTIMERS = 6,          /* # of interrupt holdoff timer values */
 	SGE_NCOUNTERS = 4,        /* # of interrupt packet counter values */
+	SGE_MAX_IQ_SIZE = 65520,
 
 	SGE_TIMER_RSTRT_CNTR = 6, /* restart RX packet threshold counter */
 	SGE_TIMER_UPD_CIDX = 7,   /* update cidx only */
diff --git a/drivers/net/ethernet/chelsio/cxgb4/t4_msg.h b/drivers/net/ethernet/chelsio/cxgb4/t4_msg.h
index f2738c71078..973eb11aa98 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/t4_msg.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/t4_msg.h
@@ -227,6 +227,7 @@ struct cpl_pass_open_req {
 #define DELACK(x)     ((x) << 5)
 #define ULP_MODE(x)   ((x) << 8)
 #define RCV_BUFSIZ(x) ((x) << 12)
+#define RCV_BUFSIZ_MASK 0x3FFU
 #define DSCP(x)       ((x) << 22)
 #define SMAC_SEL(x)   ((u64)(x) << 28)
 #define L2T_IDX(x)    ((u64)(x) << 36)
@@ -278,6 +279,15 @@ struct cpl_pass_accept_rpl {
 	__be64 opt0;
 };
 
+struct cpl_t5_pass_accept_rpl {
+	WR_HDR;
+	union opcode_tid ot;
+	__be32 opt2;
+	__be64 opt0;
+	__be32 iss;
+	__be32 rsvd;
+};
+
 struct cpl_act_open_req {
 	WR_HDR;
 	union opcode_tid ot;