ipoib_sg.patch

diff -ru mlnx-ofa_kernel-2.3/drivers/infiniband/ulp/ipoib/ipoib.h mlnx-ofa_kernel-2.3p/drivers/infiniband/ulp/ipoib/ipoib.h
--- mlnx-ofa_kernel-2.3/drivers/infiniband/ulp/ipoib/ipoib.h	2014-11-06 22:18:12.000000000 +1100
+++ mlnx-ofa_kernel-2.3p/drivers/infiniband/ulp/ipoib/ipoib.h	2014-11-18 14:59:35.509774088 +1100
@@ -658,6 +658,8 @@
 void ipoib_mcast_dev_down(struct net_device *dev);
 void ipoib_mcast_dev_flush(struct net_device *dev);
 
+int ipoib_dma_map_tx(struct ib_device *ca, struct ipoib_tx_buf *tx_req);
+
 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev);
 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter);
diff -ru mlnx-ofa_kernel-2.3/drivers/infiniband/ulp/ipoib/ipoib_cm.c mlnx-ofa_kernel-2.3p/drivers/infiniband/ulp/ipoib/ipoib_cm.c
--- mlnx-ofa_kernel-2.3/drivers/infiniband/ulp/ipoib/ipoib_cm.c	2014-11-06 22:18:08.000000000 +1100
+++ mlnx-ofa_kernel-2.3p/drivers/infiniband/ulp/ipoib/ipoib_cm.c	2014-11-18 14:51:21.513768260 +1100
@@ -732,11 +732,40 @@
 	return ib_post_send(tx->qp, &send_ring->tx_wr, &bad_wr);
 }
 
+static inline int post_send_sg(struct ipoib_dev_priv *priv,
+			       struct ipoib_cm_tx *tx,
+			       unsigned int wr_id,
+			       struct sk_buff *skb,
+			       u64 mapping[MAX_SKB_FRAGS + 1],
+			       struct ipoib_send_ring *send_ring)
+{
+	struct ib_send_wr *bad_wr;
+	int i, off;
+	skb_frag_t *frags = skb_shinfo(skb)->frags;
+	int nr_frags = skb_shinfo(skb)->nr_frags;
+	if (skb_headlen(skb)) {
+		send_ring->tx_sge[0].addr = mapping[0];
+		send_ring->tx_sge[0].length = skb_headlen(skb);
+		off = 1;
+	} else
+		off = 0;
+
+	for (i = 0; i < nr_frags; ++i) {
+		send_ring->tx_sge[i + off].addr = mapping[i + off];
+		send_ring->tx_sge[i + off].length = frags[i].size;
+	}
+	send_ring->tx_wr.num_sge = nr_frags + off;
+	send_ring->tx_wr.wr_id = wr_id | IPOIB_OP_CM;
+
+	return ib_post_send(tx->qp, &send_ring->tx_wr, &bad_wr);
+}
+
 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	struct ipoib_cm_tx_buf *tx_req;
-	u64 addr;
+	struct ipoib_tx_buf sg_tx_req;
+	u64 addr = 0;
 	int rc;
 	struct ipoib_send_ring *send_ring;
 	u16 queue_index;
@@ -768,26 +797,42 @@
 	tx_req = &tx->tx_ring[tx->tx_head & (priv->sendq_size - 1)];
 	tx_req->skb = skb;
 
-	if (skb->len < ipoib_inline_thold && !skb_shinfo(skb)->nr_frags) {
-		addr = (u64) skb->data;
-		send_ring->tx_wr.send_flags |= IB_SEND_INLINE;
-		tx_req->is_inline = 1;
-	} else {
-		addr = ib_dma_map_single(priv->ca, skb->data,
-			skb->len, DMA_TO_DEVICE);
-		if (unlikely(ib_dma_mapping_error(priv->ca, addr))) {
+	if (skb_shinfo(skb)->nr_frags) {
+		sg_tx_req.skb = skb;
+		if (unlikely(ipoib_dma_map_tx(priv->ca, &sg_tx_req))) {
 			++send_ring->stats.tx_errors;
 			dev_kfree_skb_any(skb);
 			return;
 		}
 
-		tx_req->mapping = addr;
 		tx_req->is_inline = 0;
-		send_ring->tx_wr.send_flags &= ~IB_SEND_INLINE;
+		rc = post_send_sg(priv, tx, tx->tx_head &
+				  (ipoib_sendq_size - 1),
+				  skb, sg_tx_req.mapping, send_ring);
+	} else {
+		if (skb->len < ipoib_inline_thold &&
+		    !skb_shinfo(skb)->nr_frags) {
+			addr = (u64) skb->data;
+			send_ring->tx_wr.send_flags |= IB_SEND_INLINE;
+			tx_req->is_inline = 1;
+		} else {
+			addr = ib_dma_map_single(priv->ca, skb->data,
+				skb->len, DMA_TO_DEVICE);
+			if (unlikely(ib_dma_mapping_error(priv->ca, addr))) {
+				++send_ring->stats.tx_errors;
+				dev_kfree_skb_any(skb);
+				return;
+			}
+
+			tx_req->mapping = addr;
+			tx_req->is_inline = 0;
+			send_ring->tx_wr.send_flags &= ~IB_SEND_INLINE;
+		}
+
+		rc = post_send(priv, tx, tx->tx_head & (ipoib_sendq_size - 1),
+			       addr, skb->len, send_ring);
 	}
 
-	rc = post_send(priv, tx, tx->tx_head & (priv->sendq_size - 1),
-		       addr, skb->len, send_ring);
 	if (unlikely(rc)) {
 		ipoib_warn(priv, "%s: post_send failed, error %d queue_index: %d skb->len: %d\n",
 			   __func__, rc, queue_index, skb->len);
@@ -1103,6 +1148,9 @@
 	/* CM uses ipoib_ib_completion for TX completion and work using NAPI */
 	attr.send_cq = attr.recv_cq = priv->recv_ring[tx->neigh->index].recv_cq;
 
+	if (dev->features & NETIF_F_SG)
+		attr.cap.max_send_sge = MAX_SKB_FRAGS + 1;
+
 	spin_unlock_irqrestore(&priv->lock, flags);
 
 	return ib_create_qp(priv->pd, &attr);
diff -ru mlnx-ofa_kernel-2.3/drivers/infiniband/ulp/ipoib/ipoib_ib.c mlnx-ofa_kernel-2.3p/drivers/infiniband/ulp/ipoib/ipoib_ib.c
--- mlnx-ofa_kernel-2.3/drivers/infiniband/ulp/ipoib/ipoib_ib.c	2014-11-06 22:18:08.000000000 +1100
+++ mlnx-ofa_kernel-2.3p/drivers/infiniband/ulp/ipoib/ipoib_ib.c	2014-11-18 14:57:29.930788235 +1100
@@ -414,7 +414,7 @@
 			   "for buf %d\n", wr_id);
 }
 
-static int ipoib_dma_map_tx(struct ib_device *ca,
+int ipoib_dma_map_tx(struct ib_device *ca,
 			    struct ipoib_tx_buf *tx_req)
 {
 	struct sk_buff *skb = tx_req->skb;
diff -ru mlnx-ofa_kernel-2.3/drivers/infiniband/ulp/ipoib/ipoib_main.c mlnx-ofa_kernel-2.3p/drivers/infiniband/ulp/ipoib/ipoib_main.c
--- mlnx-ofa_kernel-2.3/drivers/infiniband/ulp/ipoib/ipoib_main.c	2014-11-06 22:18:08.000000000 +1100
+++ mlnx-ofa_kernel-2.3p/drivers/infiniband/ulp/ipoib/ipoib_main.c	2014-11-18 12:47:59.780427723 +1100
@@ -201,8 +201,7 @@
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 
 	if (test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags))
-		features &= ~(NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_TSO |
-			      NETIF_F_RXCSUM);
+		features &= ~(NETIF_F_IP_CSUM | NETIF_F_TSO | NETIF_F_RXCSUM);
 
 	return features;
 }