7 files changed, 311 insertions, 148 deletions

diff --git a/drivers/infiniband/hw/cxgb4/cm.c b/drivers/infiniband/hw/cxgb4/cm.c
index a3a67216bce6..3aca7f6171b4 100644
--- a/drivers/infiniband/hw/cxgb4/cm.c
+++ b/drivers/infiniband/hw/cxgb4/cm.c
@@ -294,6 +294,25 @@ static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
 	return;
 }
 
+static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size)
+{
+	struct sk_buff *skb;
+	unsigned int i;
+	size_t len;
+
+	len = roundup(sizeof(union cpl_wr_size), 16);
+	for (i = 0; i < size; i++) {
+		skb = alloc_skb(len, GFP_KERNEL);
+		if (!skb)
+			goto fail;
+		skb_queue_tail(ep_skb_list, skb);
+	}
+	return 0;
+fail:
+	skb_queue_purge(ep_skb_list);
+	return -ENOMEM;
+}
+
 static void *alloc_ep(int size, gfp_t gfp)
 {
 	struct c4iw_ep_common *epc;
@@ -384,6 +403,8 @@ void _c4iw_free_ep(struct kref *kref)
 		if (ep->mpa_skb)
 			kfree_skb(ep->mpa_skb);
 	}
+	if (!skb_queue_empty(&ep->com.ep_skb_list))
+		skb_queue_purge(&ep->com.ep_skb_list);
 	kfree(ep);
 }
 
@@ -620,25 +641,27 @@ static void abort_arp_failure(void *handle, struct sk_buff *skb)
 	}
 }
 
-static int send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
+static int send_flowc(struct c4iw_ep *ep)
 {
-	unsigned int flowclen = 80;
 	struct fw_flowc_wr *flowc;
+	struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
 	int i;
 	u16 vlan = ep->l2t->vlan;
 	int nparams;
 
+	if (WARN_ON(!skb))
+		return -ENOMEM;
+
 	if (vlan == CPL_L2T_VLAN_NONE)
 		nparams = 8;
 	else
 		nparams = 9;
 
-	skb = get_skb(skb, flowclen, GFP_KERNEL);
-	flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
+	flowc = (struct fw_flowc_wr *)__skb_put(skb, FLOWC_LEN);
 
 	flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
 					   FW_FLOWC_WR_NPARAMS_V(nparams));
-	flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(flowclen,
+	flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(FLOWC_LEN,
 					  16)) | FW_WR_FLOWID_V(ep->hwtid));
 
 	flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
@@ -679,18 +702,16 @@ static int send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
 	return c4iw_ofld_send(&ep->com.dev->rdev, skb);
 }
 
-static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
+static int send_halfclose(struct c4iw_ep *ep)
 {
 	struct cpl_close_con_req *req;
-	struct sk_buff *skb;
+	struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
 	int wrlen = roundup(sizeof *req, 16);
 
 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
-	skb = get_skb(NULL, wrlen, gfp);
-	if (!skb) {
-		printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
+	if (WARN_ON(!skb))
 		return -ENOMEM;
-	}
+
 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
 	req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
@@ -701,26 +722,24 @@ static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
 }
 
-static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
+static int send_abort(struct c4iw_ep *ep)
 {
 	struct cpl_abort_req *req;
 	int wrlen = roundup(sizeof *req, 16);
+	struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
 
 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
-	skb = get_skb(skb, wrlen, gfp);
-	if (!skb) {
-		printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
-		       __func__);
+	if (WARN_ON(!req_skb))
 		return -ENOMEM;
-	}
-	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
-	t4_set_arp_err_handler(skb, ep, abort_arp_failure);
-	req = (struct cpl_abort_req *) skb_put(skb, wrlen);
+
+	set_wr_txq(req_skb, CPL_PRIORITY_DATA, ep->txq_idx);
+	t4_set_arp_err_handler(req_skb, ep, abort_arp_failure);
+	req = (struct cpl_abort_req *)skb_put(req_skb, wrlen);
 	memset(req, 0, wrlen);
 	INIT_TP_WR(req, ep->hwtid);
 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
 	req->cmd = CPL_ABORT_SEND_RST;
-	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
+	return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
 }
 
 static void best_mtu(const unsigned short *mtus, unsigned short mtu,
@@ -992,9 +1011,19 @@ static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
 
 	mpa = (struct mpa_message *)(req + 1);
 	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
-	mpa->flags = (crc_enabled ? MPA_CRC : 0) |
-		     (markers_enabled ? MPA_MARKERS : 0) |
-		     (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
+
+	mpa->flags = 0;
+	if (crc_enabled)
+		mpa->flags |= MPA_CRC;
+	if (markers_enabled) {
+		mpa->flags |= MPA_MARKERS;
+		ep->mpa_attr.recv_marker_enabled = 1;
+	} else {
+		ep->mpa_attr.recv_marker_enabled = 0;
+	}
+	if (mpa_rev_to_use == 2)
+		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
+
 	mpa->private_data_size = htons(ep->plen);
 	mpa->revision = mpa_rev_to_use;
 	if (mpa_rev_to_use == 1) {
@@ -1169,8 +1198,11 @@ static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
 	mpa = (struct mpa_message *)(req + 1);
 	memset(mpa, 0, sizeof(*mpa));
 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
-	mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
-		     (markers_enabled ? MPA_MARKERS : 0);
+	mpa->flags = 0;
+	if (ep->mpa_attr.crc_enabled)
+		mpa->flags |= MPA_CRC;
+	if (ep->mpa_attr.recv_marker_enabled)
+		mpa->flags |= MPA_MARKERS;
 	mpa->revision = ep->mpa_attr.version;
 	mpa->private_data_size = htons(plen);
 
@@ -1248,7 +1280,7 @@ static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
 	set_bit(ACT_ESTAB, &ep->com.history);
 
 	/* start MPA negotiation */
-	ret = send_flowc(ep, NULL);
+	ret = send_flowc(ep);
 	if (ret)
 		goto err;
 	if (ep->retry_with_mpa_v1)
@@ -1555,7 +1587,6 @@ static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
 	 */
 	__state_set(&ep->com, FPDU_MODE);
 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
-	ep->mpa_attr.recv_marker_enabled = markers_enabled;
 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
 	ep->mpa_attr.version = mpa->revision;
 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
@@ -2004,12 +2035,17 @@ static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
 }
 
 /*
- * Return whether a failed active open has allocated a TID
+ * Some of the error codes above implicitly indicate that there is no TID
+ * allocated with the result of an ACT_OPEN.  We use this predicate to make
+ * that explicit.
  */
 static inline int act_open_has_tid(int status)
 {
-	return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
-	       status != CPL_ERR_ARP_MISS;
+	return (status != CPL_ERR_TCAM_PARITY &&
+		status != CPL_ERR_TCAM_MISS &&
+		status != CPL_ERR_TCAM_FULL &&
+		status != CPL_ERR_CONN_EXIST_SYNRECV &&
+		status != CPL_ERR_CONN_EXIST);
 }
 
 /* Returns whether a CPL status conveys negative advice.
@@ -2130,6 +2166,7 @@ out:
 static int c4iw_reconnect(struct c4iw_ep *ep)
 {
 	int err = 0;
+	int size = 0;
 	struct sockaddr_in *laddr = (struct sockaddr_in *)
 				    &ep->com.cm_id->m_local_addr;
 	struct sockaddr_in *raddr = (struct sockaddr_in *)
@@ -2145,6 +2182,21 @@ static int c4iw_reconnect(struct c4iw_ep *ep)
 	init_timer(&ep->timer);
 	c4iw_init_wr_wait(&ep->com.wr_wait);
 
+	/* When MPA revision is different on nodes, the node with MPA_rev=2
+	 * tries to reconnect with MPA_rev 1 for the same EP through
+	 * c4iw_reconnect(), where the same EP is assigned with new tid for
+	 * further connection establishment. As we are using the same EP pointer
+	 * for reconnect, few skbs are used during the previous c4iw_connect(),
+	 * which leaves the EP with inadequate skbs for further
+	 * c4iw_reconnect(), Further causing an assert BUG_ON() due to empty
+	 * skb_list() during peer_abort(). Allocate skbs which is already used.
+	 */
+	size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
+	if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
+		err = -ENOMEM;
+		goto fail1;
+	}
+
 	/*
 	 * Allocate an active TID to initiate a TCP connection.
 	 */
@@ -2210,6 +2262,7 @@ fail2:
 	 * response of 1st connect request.
 	 */
 	connect_reply_upcall(ep, -ECONNRESET);
+fail1:
 	c4iw_put_ep(&ep->com);
 out:
 	return err;
@@ -2576,6 +2629,10 @@ static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
 	if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
 		child_ep->mtu = peer_mss + hdrs;
 
+	skb_queue_head_init(&child_ep->com.ep_skb_list);
+	if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
+		goto fail;
+
 	state_set(&child_ep->com, CONNECTING);
 	child_ep->com.dev = dev;
 	child_ep->com.cm_id = NULL;
@@ -2640,6 +2697,8 @@ static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
 			       (const u32 *)&sin6->sin6_addr.s6_addr, 1);
 	}
 	goto out;
+fail:
+	c4iw_put_ep(&child_ep->com);
 reject:
 	reject_cr(dev, hwtid, skb);
 	if (parent_ep)
@@ -2670,7 +2729,7 @@ static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
 	ep->com.state = MPA_REQ_WAIT;
 	start_ep_timer(ep);
 	set_bit(PASS_ESTAB, &ep->com.history);
-	ret = send_flowc(ep, skb);
+	ret = send_flowc(ep);
 	mutex_unlock(&ep->com.mutex);
 	if (ret)
 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
@@ -2871,10 +2930,8 @@ static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
 	}
 	mutex_unlock(&ep->com.mutex);
 
-	rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
-	if (!rpl_skb) {
-		printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
-		       __func__);
+	rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
+	if (WARN_ON(!rpl_skb)) {
 		release = 1;
 		goto out;
 	}
@@ -3011,9 +3068,9 @@ static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
 		PDBG("%s last streaming msg ack ep %p tid %u state %u "
 		     "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
 		     state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
+		mutex_lock(&ep->com.mutex);
 		kfree_skb(ep->mpa_skb);
 		ep->mpa_skb = NULL;
-		mutex_lock(&ep->com.mutex);
 		if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
 			stop_ep_timer(ep);
 		mutex_unlock(&ep->com.mutex);
@@ -3025,9 +3082,9 @@ out:
 
 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
 {
-	int err = 0;
-	int disconnect = 0;
+	int abort;
 	struct c4iw_ep *ep = to_ep(cm_id);
+
 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
 
 	mutex_lock(&ep->com.mutex);
@@ -3038,16 +3095,13 @@ int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
 	}
 	set_bit(ULP_REJECT, &ep->com.history);
 	if (mpa_rev == 0)
-		disconnect = 2;
-	else {
-		err = send_mpa_reject(ep, pdata, pdata_len);
-		disconnect = 1;
-	}
+		abort = 1;
+	else
+		abort = send_mpa_reject(ep, pdata, pdata_len);
 	mutex_unlock(&ep->com.mutex);
-	if (disconnect) {
-		stop_ep_timer(ep);
-		err = c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
-	}
+
+	stop_ep_timer(ep);
+	c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
 	c4iw_put_ep(&ep->com);
 	return 0;
 }
@@ -3248,6 +3302,13 @@ int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 		err = -ENOMEM;
 		goto out;
 	}
+
+	skb_queue_head_init(&ep->com.ep_skb_list);
+	if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
+		err = -ENOMEM;
+		goto fail1;
+	}
+
 	init_timer(&ep->timer);
 	ep->plen = conn_param->private_data_len;
 	if (ep->plen)
@@ -3266,7 +3327,7 @@ int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 	if (!ep->com.qp) {
 		PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
 		err = -EINVAL;
-		goto fail1;
+		goto fail2;
 	}
 	ref_qp(ep);
 	PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
@@ -3279,7 +3340,7 @@ int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 	if (ep->atid == -1) {
 		printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
 		err = -ENOMEM;
-		goto fail1;
+		goto fail2;
 	}
 	insert_handle(dev, &dev->atid_idr, ep, ep->atid);
 
@@ -3303,7 +3364,7 @@ int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 		if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
 			err = pick_local_ipaddrs(dev, cm_id);
 			if (err)
-				goto fail1;
+				goto fail2;
 		}
 
 		/* find a route */
@@ -3323,7 +3384,7 @@ int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 		if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
 			err = pick_local_ip6addrs(dev, cm_id);
 			if (err)
-				goto fail1;
+				goto fail2;
 		}
 
 		/* find a route */
@@ -3339,14 +3400,14 @@ int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 	if (!ep->dst) {
 		printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
 		err = -EHOSTUNREACH;
-		goto fail2;
+		goto fail3;
 	}
 
 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
 			ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
 	if (err) {
 		printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
-		goto fail3;
+		goto fail4;
 	}
 
 	PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
@@ -3362,13 +3423,15 @@ int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 		goto out;
 
 	cxgb4_l2t_release(ep->l2t);
-fail3:
+fail4:
 	dst_release(ep->dst);
-fail2:
+fail3:
 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
-fail1:
+fail2:
+	skb_queue_purge(&ep->com.ep_skb_list);
 	deref_cm_id(&ep->com);
+fail1:
 	c4iw_put_ep(&ep->com);
 out:
 	return err;
@@ -3461,6 +3524,7 @@ int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
 		err = -ENOMEM;
 		goto fail1;
 	}
+	skb_queue_head_init(&ep->com.ep_skb_list);
 	PDBG("%s ep %p\n", __func__, ep);
 	ep->com.cm_id = cm_id;
 	ref_cm_id(&ep->com);
@@ -3577,11 +3641,22 @@ int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
 	case MPA_REQ_RCVD:
 	case MPA_REP_SENT:
 	case FPDU_MODE:
+	case CONNECTING:
 		close = 1;
 		if (abrupt)
 			ep->com.state = ABORTING;
 		else {
 			ep->com.state = CLOSING;
+
+			/*
+			 * if we close before we see the fw4_ack() then we fix
+			 * up the timer state since we're reusing it.
+			 */
+			if (ep->mpa_skb &&
+			    test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
+				clear_bit(STOP_MPA_TIMER, &ep->com.flags);
+				stop_ep_timer(ep);
+			}
 			start_ep_timer(ep);
 		}
 		set_bit(CLOSE_SENT, &ep->com.flags);
@@ -3611,10 +3686,10 @@ int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
 		if (abrupt) {
 			set_bit(EP_DISC_ABORT, &ep->com.history);
 			close_complete_upcall(ep, -ECONNRESET);
-			ret = send_abort(ep, NULL, gfp);
+			ret = send_abort(ep);
 		} else {
 			set_bit(EP_DISC_CLOSE, &ep->com.history);
-			ret = send_halfclose(ep, gfp);
+			ret = send_halfclose(ep);
 		}
 		if (ret) {
 			set_bit(EP_DISC_FAIL, &ep->com.history);
diff --git a/drivers/infiniband/hw/cxgb4/cq.c b/drivers/infiniband/hw/cxgb4/cq.c
index b0b955724458..812ab7278b8e 100644
--- a/drivers/infiniband/hw/cxgb4/cq.c
+++ b/drivers/infiniband/hw/cxgb4/cq.c
@@ -33,19 +33,15 @@
 #include "iw_cxgb4.h"
 
 static int destroy_cq(struct c4iw_rdev *rdev, struct t4_cq *cq,
-		      struct c4iw_dev_ucontext *uctx)
+		      struct c4iw_dev_ucontext *uctx, struct sk_buff *skb)
 {
 	struct fw_ri_res_wr *res_wr;
 	struct fw_ri_res *res;
 	int wr_len;
 	struct c4iw_wr_wait wr_wait;
-	struct sk_buff *skb;
 	int ret;
 
 	wr_len = sizeof *res_wr + sizeof *res;
-	skb = alloc_skb(wr_len, GFP_KERNEL);
-	if (!skb)
-		return -ENOMEM;
 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
 
 	res_wr = (struct fw_ri_res_wr *)__skb_put(skb, wr_len);
@@ -863,7 +859,9 @@ int c4iw_destroy_cq(struct ib_cq *ib_cq)
 	ucontext = ib_cq->uobject ? to_c4iw_ucontext(ib_cq->uobject->context)
 				  : NULL;
 	destroy_cq(&chp->rhp->rdev, &chp->cq,
-		   ucontext ? &ucontext->uctx : &chp->cq.rdev->uctx);
+		   ucontext ? &ucontext->uctx : &chp->cq.rdev->uctx,
+		   chp->destroy_skb);
+	chp->destroy_skb = NULL;
 	kfree(chp);
 	return 0;
 }
@@ -879,7 +877,7 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev,
 	struct c4iw_cq *chp;
 	struct c4iw_create_cq_resp uresp;
 	struct c4iw_ucontext *ucontext = NULL;
-	int ret;
+	int ret, wr_len;
 	size_t memsize, hwentries;
 	struct c4iw_mm_entry *mm, *mm2;
 
@@ -896,6 +894,13 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev,
 	if (!chp)
 		return ERR_PTR(-ENOMEM);
 
+	wr_len = sizeof(struct fw_ri_res_wr) + sizeof(struct fw_ri_res);
+	chp->destroy_skb = alloc_skb(wr_len, GFP_KERNEL);
+	if (!chp->destroy_skb) {
+		ret = -ENOMEM;
+		goto err1;
+	}
+
 	if (ib_context)
 		ucontext = to_c4iw_ucontext(ib_context);
 
@@ -936,7 +941,7 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev,
 	ret = create_cq(&rhp->rdev, &chp->cq,
 			ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
 	if (ret)
-		goto err1;
+		goto err2;
 
 	chp->rhp = rhp;
 	chp->cq.size--;				/* status page */
@@ -947,15 +952,15 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev,
 	init_waitqueue_head(&chp->wait);
 	ret = insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid);
 	if (ret)
-		goto err2;
+		goto err3;
 
 	if (ucontext) {
 		mm = kmalloc(sizeof *mm, GFP_KERNEL);
 		if (!mm)
-			goto err3;
+			goto err4;
 		mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
 		if (!mm2)
-			goto err4;
+			goto err5;
 
 		uresp.qid_mask = rhp->rdev.cqmask;
 		uresp.cqid = chp->cq.cqid;
@@ -970,7 +975,7 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev,
 		ret = ib_copy_to_udata(udata, &uresp,
 				       sizeof(uresp) - sizeof(uresp.reserved));
 		if (ret)
-			goto err5;
+			goto err6;
 
 		mm->key = uresp.key;
 		mm->addr = virt_to_phys(chp->cq.queue);
@@ -986,15 +991,18 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev,
 	     __func__, chp->cq.cqid, chp, chp->cq.size,
 	     chp->cq.memsize, (unsigned long long) chp->cq.dma_addr);
 	return &chp->ibcq;
-err5:
+err6:
 	kfree(mm2);
-err4:
+err5:
 	kfree(mm);
-err3:
+err4:
 	remove_handle(rhp, &rhp->cqidr, chp->cq.cqid);
-err2:
+err3:
 	destroy_cq(&chp->rhp->rdev, &chp->cq,
-		   ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
+		   ucontext ? &ucontext->uctx : &rhp->rdev.uctx,
+		   chp->destroy_skb);
+err2:
+	kfree_skb(chp->destroy_skb);
 err1:
 	kfree(chp);
 	return ERR_PTR(ret);
diff --git a/drivers/infiniband/hw/cxgb4/device.c b/drivers/infiniband/hw/cxgb4/device.c
index ae2e8b23d2dd..071d7332ec06 100644
--- a/drivers/infiniband/hw/cxgb4/device.c
+++ b/drivers/infiniband/hw/cxgb4/device.c
@@ -317,7 +317,7 @@ static int qp_open(struct inode *inode, struct file *file)
 	idr_for_each(&qpd->devp->qpidr, count_idrs, &count);
 	spin_unlock_irq(&qpd->devp->lock);
 
-	qpd->bufsize = count * 128;
+	qpd->bufsize = count * 180;
 	qpd->buf = vmalloc(qpd->bufsize);
 	if (!qpd->buf) {
 		kfree(qpd);
diff --git a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
index f6f34a75af27..aa47e0ae80bc 100644
--- a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
+++ b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
@@ -384,6 +384,7 @@ struct c4iw_mr {
 	struct ib_mr ibmr;
 	struct ib_umem *umem;
 	struct c4iw_dev *rhp;
+	struct sk_buff *dereg_skb;
 	u64 kva;
 	struct tpt_attributes attr;
 	u64 *mpl;
@@ -400,6 +401,7 @@ static inline struct c4iw_mr *to_c4iw_mr(struct ib_mr *ibmr)
 struct c4iw_mw {
 	struct ib_mw ibmw;
 	struct c4iw_dev *rhp;
+	struct sk_buff *dereg_skb;
 	u64 kva;
 	struct tpt_attributes attr;
 };
@@ -412,6 +414,7 @@ static inline struct c4iw_mw *to_c4iw_mw(struct ib_mw *ibmw)
 struct c4iw_cq {
 	struct ib_cq ibcq;
 	struct c4iw_dev *rhp;
+	struct sk_buff *destroy_skb;
 	struct t4_cq cq;
 	spinlock_t lock;
 	spinlock_t comp_handler_lock;
@@ -472,7 +475,7 @@ struct c4iw_qp {
 	struct t4_wq wq;
 	spinlock_t lock;
 	struct mutex mutex;
-	atomic_t refcnt;
+	struct kref kref;
 	wait_queue_head_t wait;
 	struct timer_list timer;
 	int sq_sig_all;
@@ -789,10 +792,29 @@ enum c4iw_ep_history {
 	CM_ID_DEREFED		= 28,
 };
 
+enum conn_pre_alloc_buffers {
+	CN_ABORT_REQ_BUF,
+	CN_ABORT_RPL_BUF,
+	CN_CLOSE_CON_REQ_BUF,
+	CN_DESTROY_BUF,
+	CN_FLOWC_BUF,
+	CN_MAX_CON_BUF
+};
+
+#define FLOWC_LEN 80
+union cpl_wr_size {
+	struct cpl_abort_req abrt_req;
+	struct cpl_abort_rpl abrt_rpl;
+	struct fw_ri_wr ri_req;
+	struct cpl_close_con_req close_req;
+	char flowc_buf[FLOWC_LEN];
+};
+
 struct c4iw_ep_common {
 	struct iw_cm_id *cm_id;
 	struct c4iw_qp *qp;
 	struct c4iw_dev *dev;
+	struct sk_buff_head ep_skb_list;
 	enum c4iw_ep_state state;
 	struct kref kref;
 	struct mutex mutex;
diff --git a/drivers/infiniband/hw/cxgb4/mem.c b/drivers/infiniband/hw/cxgb4/mem.c
index 55d0651ee4de..0b91b0f4df71 100644
--- a/drivers/infiniband/hw/cxgb4/mem.c
+++ b/drivers/infiniband/hw/cxgb4/mem.c
@@ -59,9 +59,9 @@ static int mr_exceeds_hw_limits(struct c4iw_dev *dev, u64 length)
 }
 
 static int _c4iw_write_mem_dma_aligned(struct c4iw_rdev *rdev, u32 addr,
-				       u32 len, dma_addr_t data, int wait)
+				       u32 len, dma_addr_t data,
+				       int wait, struct sk_buff *skb)
 {
-	struct sk_buff *skb;
 	struct ulp_mem_io *req;
 	struct ulptx_sgl *sgl;
 	u8 wr_len;
@@ -74,9 +74,11 @@ static int _c4iw_write_mem_dma_aligned(struct c4iw_rdev *rdev, u32 addr,
 		c4iw_init_wr_wait(&wr_wait);
 	wr_len = roundup(sizeof(*req) + sizeof(*sgl), 16);
 
-	skb = alloc_skb(wr_len, GFP_KERNEL);
-	if (!skb)
-		return -ENOMEM;
+	if (!skb) {
+		skb = alloc_skb(wr_len, GFP_KERNEL | __GFP_NOFAIL);
+		if (!skb)
+			return -ENOMEM;
+	}
 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
 
 	req = (struct ulp_mem_io *)__skb_put(skb, wr_len);
@@ -108,9 +110,8 @@ static int _c4iw_write_mem_dma_aligned(struct c4iw_rdev *rdev, u32 addr,
 }
 
 static int _c4iw_write_mem_inline(struct c4iw_rdev *rdev, u32 addr, u32 len,
-				  void *data)
+				  void *data, struct sk_buff *skb)
 {
-	struct sk_buff *skb;
 	struct ulp_mem_io *req;
 	struct ulptx_idata *sc;
 	u8 wr_len, *to_dp, *from_dp;
@@ -134,9 +135,11 @@ static int _c4iw_write_mem_inline(struct c4iw_rdev *rdev, u32 addr, u32 len,
 		wr_len = roundup(sizeof *req + sizeof *sc +
 				 roundup(copy_len, T4_ULPTX_MIN_IO), 16);
 
-		skb = alloc_skb(wr_len, GFP_KERNEL);
-		if (!skb)
-			return -ENOMEM;
+		if (!skb) {
+			skb = alloc_skb(wr_len, GFP_KERNEL | __GFP_NOFAIL);
+			if (!skb)
+				return -ENOMEM;
+		}
 		set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
 
 		req = (struct ulp_mem_io *)__skb_put(skb, wr_len);
@@ -173,6 +176,7 @@ static int _c4iw_write_mem_inline(struct c4iw_rdev *rdev, u32 addr, u32 len,
 			memset(to_dp + copy_len, 0, T4_ULPTX_MIN_IO -
 			       (copy_len % T4_ULPTX_MIN_IO));
 		ret = c4iw_ofld_send(rdev, skb);
+		skb = NULL;
 		if (ret)
 			return ret;
 		len -= C4IW_MAX_INLINE_SIZE;
@@ -182,7 +186,8 @@ static int _c4iw_write_mem_inline(struct c4iw_rdev *rdev, u32 addr, u32 len,
 	return ret;
 }
 
-static int _c4iw_write_mem_dma(struct c4iw_rdev *rdev, u32 addr, u32 len, void *data)
+static int _c4iw_write_mem_dma(struct c4iw_rdev *rdev, u32 addr, u32 len,
+			       void *data, struct sk_buff *skb)
 {
 	u32 remain = len;
 	u32 dmalen;
@@ -205,7 +210,7 @@ static int _c4iw_write_mem_dma(struct c4iw_rdev *rdev, u32 addr, u32 len, void *
 			dmalen = T4_ULPTX_MAX_DMA;
 		remain -= dmalen;
 		ret = _c4iw_write_mem_dma_aligned(rdev, addr, dmalen, daddr,
-						 !remain);
+						 !remain, skb);
 		if (ret)
 			goto out;
 		addr += dmalen >> 5;
@@ -213,7 +218,7 @@ static int _c4iw_write_mem_dma(struct c4iw_rdev *rdev, u32 addr, u32 len, void *
 		daddr += dmalen;
 	}
 	if (remain)
-		ret = _c4iw_write_mem_inline(rdev, addr, remain, data);
+		ret = _c4iw_write_mem_inline(rdev, addr, remain, data, skb);
 out:
 	dma_unmap_single(&rdev->lldi.pdev->dev, save, len, DMA_TO_DEVICE);
 	return ret;
@@ -224,23 +229,25 @@ out:
  * If data is NULL, clear len byte of memory to zero.
  */
 static int write_adapter_mem(struct c4iw_rdev *rdev, u32 addr, u32 len,
-			     void *data)
+			     void *data, struct sk_buff *skb)
 {
 	if (is_t5(rdev->lldi.adapter_type) && use_dsgl) {
 		if (len > inline_threshold) {
-			if (_c4iw_write_mem_dma(rdev, addr, len, data)) {
+			if (_c4iw_write_mem_dma(rdev, addr, len, data, skb)) {
 				printk_ratelimited(KERN_WARNING
 						   "%s: dma map"
 						   " failure (non fatal)\n",
 						   pci_name(rdev->lldi.pdev));
 				return _c4iw_write_mem_inline(rdev, addr, len,
-							      data);
-			} else
+							      data, skb);
+			} else {
 				return 0;
+			}
 		} else
-			return _c4iw_write_mem_inline(rdev, addr, len, data);
+			return _c4iw_write_mem_inline(rdev, addr,
+						      len, data, skb);
 	} else
-		return _c4iw_write_mem_inline(rdev, addr, len, data);
+		return _c4iw_write_mem_inline(rdev, addr, len, data, skb);
 }
 
 /*
@@ -253,7 +260,8 @@ static int write_tpt_entry(struct c4iw_rdev *rdev, u32 reset_tpt_entry,
 			   u32 *stag, u8 stag_state, u32 pdid,
 			   enum fw_ri_stag_type type, enum fw_ri_mem_perms perm,
 			   int bind_enabled, u32 zbva, u64 to,
-			   u64 len, u8 page_size, u32 pbl_size, u32 pbl_addr)
+			   u64 len, u8 page_size, u32 pbl_size, u32 pbl_addr,
+			   struct sk_buff *skb)
 {
 	int err;
 	struct fw_ri_tpte tpt;
@@ -307,7 +315,7 @@ static int write_tpt_entry(struct c4iw_rdev *rdev, u32 reset_tpt_entry,
 	}
 	err = write_adapter_mem(rdev, stag_idx +
 				(rdev->lldi.vr->stag.start >> 5),
-				sizeof(tpt), &tpt);
+				sizeof(tpt), &tpt, skb);
 
 	if (reset_tpt_entry) {
 		c4iw_put_resource(&rdev->resource.tpt_table, stag_idx);
@@ -327,28 +335,29 @@ static int write_pbl(struct c4iw_rdev *rdev, __be64 *pbl,
 	     __func__, pbl_addr, rdev->lldi.vr->pbl.start,
 	     pbl_size);
 
-	err = write_adapter_mem(rdev, pbl_addr >> 5, pbl_size << 3, pbl);
+	err = write_adapter_mem(rdev, pbl_addr >> 5, pbl_size << 3, pbl, NULL);
 	return err;
 }
 
 static int dereg_mem(struct c4iw_rdev *rdev, u32 stag, u32 pbl_size,
-		     u32 pbl_addr)
+		     u32 pbl_addr, struct sk_buff *skb)
 {
 	return write_tpt_entry(rdev, 1, &stag, 0, 0, 0, 0, 0, 0, 0UL, 0, 0,
-			       pbl_size, pbl_addr);
+			       pbl_size, pbl_addr, skb);
 }
 
 static int allocate_window(struct c4iw_rdev *rdev, u32 * stag, u32 pdid)
 {
 	*stag = T4_STAG_UNSET;
 	return write_tpt_entry(rdev, 0, stag, 0, pdid, FW_RI_STAG_MW, 0, 0, 0,
-			       0UL, 0, 0, 0, 0);
+			       0UL, 0, 0, 0, 0, NULL);
 }
 
-static int deallocate_window(struct c4iw_rdev *rdev, u32 stag)
+static int deallocate_window(struct c4iw_rdev *rdev, u32 stag,
+			     struct sk_buff *skb)
 {
 	return write_tpt_entry(rdev, 1, &stag, 0, 0, 0, 0, 0, 0, 0UL, 0, 0, 0,
-			       0);
+			       0, skb);
 }
 
 static int allocate_stag(struct c4iw_rdev *rdev, u32 *stag, u32 pdid,
@@ -356,7 +365,7 @@ static int allocate_stag(struct c4iw_rdev *rdev, u32 *stag, u32 pdid,
 {
 	*stag = T4_STAG_UNSET;
 	return write_tpt_entry(rdev, 0, stag, 0, pdid, FW_RI_STAG_NSMR, 0, 0, 0,
-			       0UL, 0, 0, pbl_size, pbl_addr);
+			       0UL, 0, 0, pbl_size, pbl_addr, NULL);
 }
 
 static int finish_mem_reg(struct c4iw_mr *mhp, u32 stag)
@@ -383,14 +392,16 @@ static int register_mem(struct c4iw_dev *rhp, struct c4iw_pd *php,
 			      mhp->attr.mw_bind_enable, mhp->attr.zbva,
 			      mhp->attr.va_fbo, mhp->attr.len ?
 			      mhp->attr.len : -1, shift - 12,
-			      mhp->attr.pbl_size, mhp->attr.pbl_addr);
+			      mhp->attr.pbl_size, mhp->attr.pbl_addr, NULL);
 	if (ret)
 		return ret;
 
 	ret = finish_mem_reg(mhp, stag);
-	if (ret)
+	if (ret) {
 		dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
-		       mhp->attr.pbl_addr);
+			  mhp->attr.pbl_addr, mhp->dereg_skb);
+		mhp->dereg_skb = NULL;
+	}
 	return ret;
 }
 
@@ -423,6 +434,12 @@ struct ib_mr *c4iw_get_dma_mr(struct ib_pd *pd, int acc)
 	if (!mhp)
 		return ERR_PTR(-ENOMEM);
 
+	mhp->dereg_skb = alloc_skb(SGE_MAX_WR_LEN, GFP_KERNEL);
+	if (!mhp->dereg_skb) {
+		ret = -ENOMEM;
+		goto err0;
+	}
+
 	mhp->rhp = rhp;
 	mhp->attr.pdid = php->pdid;
 	mhp->attr.perms = c4iw_ib_to_tpt_access(acc);
@@ -435,7 +452,8 @@ struct ib_mr *c4iw_get_dma_mr(struct ib_pd *pd, int acc)
 
 	ret = write_tpt_entry(&rhp->rdev, 0, &stag, 1, php->pdid,
 			      FW_RI_STAG_NSMR, mhp->attr.perms,
-			      mhp->attr.mw_bind_enable, 0, 0, ~0ULL, 0, 0, 0);
+			      mhp->attr.mw_bind_enable, 0, 0, ~0ULL, 0, 0, 0,
+			      NULL);
 	if (ret)
 		goto err1;
 
@@ -445,8 +463,10 @@ struct ib_mr *c4iw_get_dma_mr(struct ib_pd *pd, int acc)
 	return &mhp->ibmr;
 err2:
 	dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
-		  mhp->attr.pbl_addr);
+		  mhp->attr.pbl_addr, mhp->dereg_skb);
 err1:
+	kfree_skb(mhp->dereg_skb);
+err0:
 	kfree(mhp);
 	return ERR_PTR(ret);
 }
@@ -481,11 +501,18 @@ struct ib_mr *c4iw_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	if (!mhp)
 		return ERR_PTR(-ENOMEM);
 
+	mhp->dereg_skb = alloc_skb(SGE_MAX_WR_LEN, GFP_KERNEL);
+	if (!mhp->dereg_skb) {
+		kfree(mhp);
+		return ERR_PTR(-ENOMEM);
+	}
+
 	mhp->rhp = rhp;
 
 	mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0);
 	if (IS_ERR(mhp->umem)) {
 		err = PTR_ERR(mhp->umem);
+		kfree_skb(mhp->dereg_skb);
 		kfree(mhp);
 		return ERR_PTR(err);
 	}
@@ -550,6 +577,7 @@ err_pbl:
 
 err:
 	ib_umem_release(mhp->umem);
+	kfree_skb(mhp->dereg_skb);
 	kfree(mhp);
 	return ERR_PTR(err);
 }
@@ -572,11 +600,16 @@ struct ib_mw *c4iw_alloc_mw(struct ib_pd *pd, enum ib_mw_type type,
 	mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 	if (!mhp)
 		return ERR_PTR(-ENOMEM);
-	ret = allocate_window(&rhp->rdev, &stag, php->pdid);
-	if (ret) {
-		kfree(mhp);
-		return ERR_PTR(ret);
+
+	mhp->dereg_skb = alloc_skb(SGE_MAX_WR_LEN, GFP_KERNEL);
+	if (!mhp->dereg_skb) {
+		ret = -ENOMEM;
+		goto free_mhp;
 	}
+
+	ret = allocate_window(&rhp->rdev, &stag, php->pdid);
+	if (ret)
+		goto free_skb;
 	mhp->rhp = rhp;
 	mhp->attr.pdid = php->pdid;
 	mhp->attr.type = FW_RI_STAG_MW;
@@ -584,12 +617,19 @@ struct ib_mw *c4iw_alloc_mw(struct ib_pd *pd, enum ib_mw_type type,
 	mmid = (stag) >> 8;
 	mhp->ibmw.rkey = stag;
 	if (insert_handle(rhp, &rhp->mmidr, mhp, mmid)) {
-		deallocate_window(&rhp->rdev, mhp->attr.stag);
-		kfree(mhp);
-		return ERR_PTR(-ENOMEM);
+		ret = -ENOMEM;
+		goto dealloc_win;
 	}
 	PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 	return &(mhp->ibmw);
+
+dealloc_win:
+	deallocate_window(&rhp->rdev, mhp->attr.stag, mhp->dereg_skb);
+free_skb:
+	kfree_skb(mhp->dereg_skb);
+free_mhp:
+	kfree(mhp);
+	return ERR_PTR(ret);
 }
 
 int c4iw_dealloc_mw(struct ib_mw *mw)
@@ -602,7 +642,8 @@ int c4iw_dealloc_mw(struct ib_mw *mw)
 	rhp = mhp->rhp;
 	mmid = (mw->rkey) >> 8;
 	remove_handle(rhp, &rhp->mmidr, mmid);
-	deallocate_window(&rhp->rdev, mhp->attr.stag);
+	deallocate_window(&rhp->rdev, mhp->attr.stag, mhp->dereg_skb);
+	kfree_skb(mhp->dereg_skb);
 	kfree(mhp);
 	PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __func__, mw, mmid, mhp);
 	return 0;
@@ -666,7 +707,7 @@ struct ib_mr *c4iw_alloc_mr(struct ib_pd *pd,
 	return &(mhp->ibmr);
 err3:
 	dereg_mem(&rhp->rdev, stag, mhp->attr.pbl_size,
-		       mhp->attr.pbl_addr);
+		  mhp->attr.pbl_addr, mhp->dereg_skb);
 err2:
 	c4iw_pblpool_free(&mhp->rhp->rdev, mhp->attr.pbl_addr,
 			      mhp->attr.pbl_size << 3);
@@ -717,7 +758,7 @@ int c4iw_dereg_mr(struct ib_mr *ib_mr)
 		dma_free_coherent(&mhp->rhp->rdev.lldi.pdev->dev,
 				  mhp->max_mpl_len, mhp->mpl, mhp->mpl_addr);
 	dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
-		       mhp->attr.pbl_addr);
+		  mhp->attr.pbl_addr, mhp->dereg_skb);
 	if (mhp->attr.pbl_size)
 		c4iw_pblpool_free(&mhp->rhp->rdev, mhp->attr.pbl_addr,
 				  mhp->attr.pbl_size << 3);
diff --git a/drivers/infiniband/hw/cxgb4/provider.c b/drivers/infiniband/hw/cxgb4/provider.c
index dd8a86b726d2..df127ce6b6ec 100644
--- a/drivers/infiniband/hw/cxgb4/provider.c
+++ b/drivers/infiniband/hw/cxgb4/provider.c
@@ -409,20 +409,6 @@ static ssize_t show_rev(struct device *dev, struct device_attribute *attr,
 		       CHELSIO_CHIP_RELEASE(c4iw_dev->rdev.lldi.adapter_type));
 }
 
-static ssize_t show_fw_ver(struct device *dev, struct device_attribute *attr,
-			   char *buf)
-{
-	struct c4iw_dev *c4iw_dev = container_of(dev, struct c4iw_dev,
-						 ibdev.dev);
-	PDBG("%s dev 0x%p\n", __func__, dev);
-
-	return sprintf(buf, "%u.%u.%u.%u\n",
-			FW_HDR_FW_VER_MAJOR_G(c4iw_dev->rdev.lldi.fw_vers),
-			FW_HDR_FW_VER_MINOR_G(c4iw_dev->rdev.lldi.fw_vers),
-			FW_HDR_FW_VER_MICRO_G(c4iw_dev->rdev.lldi.fw_vers),
-			FW_HDR_FW_VER_BUILD_G(c4iw_dev->rdev.lldi.fw_vers));
-}
-
 static ssize_t show_hca(struct device *dev, struct device_attribute *attr,
 			char *buf)
 {
@@ -502,13 +488,11 @@ static int c4iw_get_mib(struct ib_device *ibdev,
 }
 
 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
-static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
 
 static struct device_attribute *c4iw_class_attributes[] = {
 	&dev_attr_hw_rev,
-	&dev_attr_fw_ver,
 	&dev_attr_hca_type,
 	&dev_attr_board_id,
 };
@@ -530,6 +514,20 @@ static int c4iw_port_immutable(struct ib_device *ibdev, u8 port_num,
 	return 0;
 }
 
+static void get_dev_fw_str(struct ib_device *dev, char *str,
+			   size_t str_len)
+{
+	struct c4iw_dev *c4iw_dev = container_of(dev, struct c4iw_dev,
+						 ibdev);
+	PDBG("%s dev 0x%p\n", __func__, dev);
+
+	snprintf(str, str_len, "%u.%u.%u.%u",
+		 FW_HDR_FW_VER_MAJOR_G(c4iw_dev->rdev.lldi.fw_vers),
+		 FW_HDR_FW_VER_MINOR_G(c4iw_dev->rdev.lldi.fw_vers),
+		 FW_HDR_FW_VER_MICRO_G(c4iw_dev->rdev.lldi.fw_vers),
+		 FW_HDR_FW_VER_BUILD_G(c4iw_dev->rdev.lldi.fw_vers));
+}
+
 int c4iw_register_device(struct c4iw_dev *dev)
 {
 	int ret;
@@ -605,6 +603,7 @@ int c4iw_register_device(struct c4iw_dev *dev)
 	dev->ibdev.get_hw_stats = c4iw_get_mib;
 	dev->ibdev.uverbs_abi_ver = C4IW_UVERBS_ABI_VERSION;
 	dev->ibdev.get_port_immutable = c4iw_port_immutable;
+	dev->ibdev.get_dev_fw_str = get_dev_fw_str;
 	dev->ibdev.drain_sq = c4iw_drain_sq;
 	dev->ibdev.drain_rq = c4iw_drain_rq;
 
diff --git a/drivers/infiniband/hw/cxgb4/qp.c b/drivers/infiniband/hw/cxgb4/qp.c
index e8993e49b8b3..edb1172b6f54 100644
--- a/drivers/infiniband/hw/cxgb4/qp.c
+++ b/drivers/infiniband/hw/cxgb4/qp.c
@@ -683,17 +683,25 @@ static int build_inv_stag(union t4_wr *wqe, struct ib_send_wr *wr,
 	return 0;
 }
 
+void _free_qp(struct kref *kref)
+{
+	struct c4iw_qp *qhp;
+
+	qhp = container_of(kref, struct c4iw_qp, kref);
+	PDBG("%s qhp %p\n", __func__, qhp);
+	kfree(qhp);
+}
+
 void c4iw_qp_add_ref(struct ib_qp *qp)
 {
 	PDBG("%s ib_qp %p\n", __func__, qp);
-	atomic_inc(&(to_c4iw_qp(qp)->refcnt));
+	kref_get(&to_c4iw_qp(qp)->kref);
 }
 
 void c4iw_qp_rem_ref(struct ib_qp *qp)
 {
 	PDBG("%s ib_qp %p\n", __func__, qp);
-	if (atomic_dec_and_test(&(to_c4iw_qp(qp)->refcnt)))
-		wake_up(&(to_c4iw_qp(qp)->wait));
+	kref_put(&to_c4iw_qp(qp)->kref, _free_qp);
 }
 
 static void add_to_fc_list(struct list_head *head, struct list_head *entry)
@@ -1081,9 +1089,10 @@ static void post_terminate(struct c4iw_qp *qhp, struct t4_cqe *err_cqe,
 	PDBG("%s qhp %p qid 0x%x tid %u\n", __func__, qhp, qhp->wq.sq.qid,
 	     qhp->ep->hwtid);
 
-	skb = alloc_skb(sizeof *wqe, gfp);
-	if (!skb)
+	skb = skb_dequeue(&qhp->ep->com.ep_skb_list);
+	if (WARN_ON(!skb))
 		return;
+
 	set_wr_txq(skb, CPL_PRIORITY_DATA, qhp->ep->txq_idx);
 
 	wqe = (struct fw_ri_wr *)__skb_put(skb, sizeof(*wqe));
@@ -1202,9 +1211,10 @@ static int rdma_fini(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
 	PDBG("%s qhp %p qid 0x%x tid %u\n", __func__, qhp, qhp->wq.sq.qid,
 	     ep->hwtid);
 
-	skb = alloc_skb(sizeof *wqe, GFP_KERNEL);
-	if (!skb)
+	skb = skb_dequeue(&ep->com.ep_skb_list);
+	if (WARN_ON(!skb))
 		return -ENOMEM;
+
 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
 
 	wqe = (struct fw_ri_wr *)__skb_put(skb, sizeof(*wqe));
@@ -1592,8 +1602,6 @@ int c4iw_destroy_qp(struct ib_qp *ib_qp)
 	wait_event(qhp->wait, !qhp->ep);
 
 	remove_handle(rhp, &rhp->qpidr, qhp->wq.sq.qid);
-	atomic_dec(&qhp->refcnt);
-	wait_event(qhp->wait, !atomic_read(&qhp->refcnt));
 
 	spin_lock_irq(&rhp->lock);
 	if (!list_empty(&qhp->db_fc_entry))
@@ -1606,8 +1614,9 @@ int c4iw_destroy_qp(struct ib_qp *ib_qp)
 	destroy_qp(&rhp->rdev, &qhp->wq,
 		   ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
 
+	c4iw_qp_rem_ref(ib_qp);
+
 	PDBG("%s ib_qp %p qpid 0x%0x\n", __func__, ib_qp, qhp->wq.sq.qid);
-	kfree(qhp);
 	return 0;
 }
 
@@ -1704,7 +1713,7 @@ struct ib_qp *c4iw_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attrs,
 	init_completion(&qhp->rq_drained);
 	mutex_init(&qhp->mutex);
 	init_waitqueue_head(&qhp->wait);
-	atomic_set(&qhp->refcnt, 1);
+	kref_init(&qhp->kref);
 
 	ret = insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.sq.qid);
 	if (ret)
@@ -1896,12 +1905,20 @@ int c4iw_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 	return 0;
 }
 
+static void move_qp_to_err(struct c4iw_qp *qp)
+{
+	struct c4iw_qp_attributes attrs = { .next_state = C4IW_QP_STATE_ERROR };
+
+	(void)c4iw_modify_qp(qp->rhp, qp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
+}
+
 void c4iw_drain_sq(struct ib_qp *ibqp)
 {
 	struct c4iw_qp *qp = to_c4iw_qp(ibqp);
 	unsigned long flag;
 	bool need_to_wait;
 
+	move_qp_to_err(qp);
 	spin_lock_irqsave(&qp->lock, flag);
 	need_to_wait = !t4_sq_empty(&qp->wq);
 	spin_unlock_irqrestore(&qp->lock, flag);
@@ -1916,6 +1933,7 @@ void c4iw_drain_rq(struct ib_qp *ibqp)
 	unsigned long flag;
 	bool need_to_wait;
 
+	move_qp_to_err(qp);
 	spin_lock_irqsave(&qp->lock, flag);
 	need_to_wait = !t4_rq_empty(&qp->wq);
 	spin_unlock_irqrestore(&qp->lock, flag);