[ofa-general] [PATCH] IB/ipoib: Add checksum offload support

[Eli Cohen]

IB/ipoib: Add checksum offload support

If the HCA supports checksum offload it will have IB_DEVICE_UD_IP_CSUM set in
its capability flags and this will cause ipoib to set NETIF_F_IP_CSUM. It also
means that the HCA can verify checksum of incoming packets.

Signed-off-by: Eli Cohen <eli at mellanox.co.il>
---
 drivers/infiniband/ulp/ipoib/ipoib.h      |    2 +
 drivers/infiniband/ulp/ipoib/ipoib_cm.c   |    7 ++++++
 drivers/infiniband/ulp/ipoib/ipoib_ib.c   |   12 ++++++++++
 drivers/infiniband/ulp/ipoib/ipoib_main.c |   33 +++++++++++++++++++++++++++++
 4 files changed, 54 insertions(+), 0 deletions(-)

diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h
index 054fab8..19a41ff 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib.h
+++ b/drivers/infiniband/ulp/ipoib/ipoib.h
@@ -87,6 +87,7 @@ enum {
 	IPOIB_MCAST_STARTED	  = 8,
 	IPOIB_FLAG_ADMIN_CM	  = 9,
 	IPOIB_FLAG_UMCAST	  = 10,
+	IPOIB_FLAG_CSUM		  = 11,
 
 	IPOIB_MAX_BACKOFF_SECONDS = 16,
 
@@ -318,6 +319,7 @@ struct ipoib_dev_priv {
 	struct dentry *mcg_dentry;
 	struct dentry *path_dentry;
 #endif
+	int	hca_caps;
 };
 
 struct ipoib_ah {
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_cm.c b/drivers/infiniband/ulp/ipoib/ipoib_cm.c
index 52b1beb..3bdb4dd 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_cm.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_cm.c
@@ -1380,6 +1380,9 @@ static ssize_t set_mode(struct device *d, struct device_attribute *attr,
 		set_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
 		ipoib_warn(priv, "enabling connected mode "
 			   "will cause multicast packet drops\n");
+
+		dev->features &= ~(NETIF_F_IP_CSUM | NETIF_F_SG);
+
 		ipoib_flush_paths(dev);
 		return count;
 	}
@@ -1388,6 +1391,10 @@ static ssize_t set_mode(struct device *d, struct device_attribute *attr,
 		clear_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
 		dev->mtu = min(priv->mcast_mtu, dev->mtu);
 		ipoib_flush_paths(dev);
+
+		if (priv->hca_caps & IB_DEVICE_UD_IP_CSUM)
+			dev->features |= NETIF_F_IP_CSUM | NETIF_F_SG;
+
 		return count;
 	}
 
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
index 08c4396..7450ce0 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
@@ -37,6 +37,7 @@
 
 #include <linux/delay.h>
 #include <linux/dma-mapping.h>
+#include <linux/ip.h>
 
 #include <rdma/ib_cache.h>
 
@@ -231,6 +232,11 @@ static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
 	skb->dev = dev;
 	/* XXX get correct PACKET_ type here */
 	skb->pkt_type = PACKET_HOST;
+
+	/* check rx csum */
+	if (test_bit(IPOIB_FLAG_CSUM, &priv->flags) && likely(wc->csum_ok))
+			skb->ip_summed = CHECKSUM_UNNECESSARY;
+
 	netif_receive_skb(skb);
 
 repost:
@@ -442,6 +448,12 @@ void ipoib_send(struct net_device *dev, struct sk_buff *skb,
 		return;
 	}
 
+	if (dev->flags & NETIF_F_IP_CSUM &&
+	    skb->ip_summed == CHECKSUM_PARTIAL)
+		priv->tx_wr.send_flags |= IB_SEND_IP_CSUM;
+	else
+		priv->tx_wr.send_flags &= ~IB_SEND_IP_CSUM;
+
 	if (unlikely(post_send(priv, priv->tx_head & (ipoib_sendq_size - 1),
 			       address->ah, qpn,
 			       tx_req->mapping, skb_headlen(skb),
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c
index f96477a..2c2a922 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_main.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c
@@ -1100,11 +1100,26 @@ int ipoib_add_pkey_attr(struct net_device *dev)
 	return device_create_file(&dev->dev, &dev_attr_pkey);
 }
 
+static void set_csum(struct net_device *dev)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+
+	if (test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags))
+		return;
+
+	if (!(priv->hca_caps & IB_DEVICE_UD_IP_CSUM))
+		return;
+
+	dev->features |= NETIF_F_SG | NETIF_F_IP_CSUM;
+	set_bit(IPOIB_FLAG_CSUM, &priv->flags);
+}
+
 static struct net_device *ipoib_add_port(const char *format,
 					 struct ib_device *hca, u8 port)
 {
 	struct ipoib_dev_priv *priv;
 	int result = -ENOMEM;
+	struct ib_device_attr *device_attr;
 
 	priv = ipoib_intf_alloc(format);
 	if (!priv)
@@ -1119,6 +1134,23 @@ static struct net_device *ipoib_add_port(const char *format,
 		goto device_init_failed;
 	}
 
+	device_attr = kmalloc(sizeof *device_attr, GFP_KERNEL);
+	if (!device_attr) {
+		printk(KERN_WARNING "%s: allocation of %lu bytes failed\n",
+		       hca->name, sizeof *device_attr);
+		goto device_init_failed;
+	}
+
+	result = ib_query_device(hca, device_attr);
+	if (result) {
+		printk(KERN_WARNING "%s: ib_query_pkey port %d failed (ret = %d)\n",
+		       hca->name, port, result);
+		kfree(device_attr);
+		goto device_init_failed;
+	}
+	priv->hca_caps = device_attr->device_cap_flags;
+	kfree(device_attr);
+
 	/*
 	 * Set the full membership bit, so that we join the right
 	 * broadcast group, etc.
@@ -1136,6 +1168,7 @@ static struct net_device *ipoib_add_port(const char *format,
 	} else
 		memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
 
+	set_csum(priv->dev);
 
 	result = ipoib_dev_init(priv->dev, hca, port);
 	if (result < 0) {
-- 
1.5.4.2