IPoIB: Connected mode experimental support

The following patch adds experimental support for IPoIB connected
mode, as defined by the draft from the IETF ipoib working group.  The
idea is to increase performance by increasing the MTU from the maximum
of 2K (theoretically 4K) supported by IPoIB on top of UD.  With this
code, I'm able to get 800MByte/sec or more with netperf without
options on a Mellanox 4x back-to-back DDR system.

Some notes on code:
1. SRQ is used for scalability to large cluster sizes
2. Only RC connections are used (UC does not support SRQ now)
3. Retry count is set to 0 since spec draft warns against retries
4. Each connection is used for data transfers in only 1 direction, so
   each connection is either active(TX) or passive (RX).  2 sides that
   want to communicate create 2 connections.
5. Each active (TX) connection has a separate CQ for send completions -
   this keeps the code simple without CQ resize and other tricks
6. To detect stale passive side connections (where the remote side is
   down), we keep an LRU list of passive connections (updated once per
   second per connection) and destroy a connection after it has been
   unused for several seconds. The LRU rule makes it possible to avoid
   scanning connections that have recently been active.

Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

1 files changed, 58 insertions, 5 deletions

diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c
index af5ee2e..18d27fd 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_main.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c
@@ -49,8 +49,6 @@
 
 #include <net/dst.h>
 
-#define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
-
 MODULE_AUTHOR("Roland Dreier");
 MODULE_DESCRIPTION("IP-over-InfiniBand net driver");
 MODULE_LICENSE("Dual BSD/GPL");
@@ -145,6 +143,8 @@ static int ipoib_stop(struct net_device *dev)
 
 	netif_stop_queue(dev);
 
+	clear_bit(IPOIB_FLAG_NETIF_STOPPED, &priv->flags);
+
 	/*
 	 * Now flush workqueue to make sure a scheduled task doesn't
 	 * bring our internal state back up.
@@ -178,8 +178,18 @@ static int ipoib_change_mtu(struct net_device *dev, int new_mtu)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 
-	if (new_mtu > IPOIB_PACKET_SIZE - IPOIB_ENCAP_LEN)
+	/* dev->mtu > 2K ==> connected mode */
+	if (ipoib_cm_admin_enabled(dev) && new_mtu <= IPOIB_CM_MTU) {
+		if (new_mtu > priv->mcast_mtu)
+			ipoib_warn(priv, "mtu > %d will cause multicast packet drops.\n",
+				   priv->mcast_mtu);
+		dev->mtu = new_mtu;
+		return 0;
+	}
+
+	if (new_mtu > IPOIB_PACKET_SIZE - IPOIB_ENCAP_LEN) {
 		return -EINVAL;
+	}
 
 	priv->admin_mtu = new_mtu;
 
@@ -414,6 +424,20 @@ static void path_rec_completion(int status,
 			memcpy(&neigh->dgid.raw, &path->pathrec.dgid.raw,
 			       sizeof(union ib_gid));
 
+			if (ipoib_cm_enabled(dev, neigh->neighbour)) {
+				if (!ipoib_cm_get(neigh))
+					ipoib_cm_set(neigh, ipoib_cm_create_tx(dev,
+									       path,
+									       neigh));
+				if (!ipoib_cm_get(neigh)) {
+					list_del(&neigh->list);
+					if (neigh->ah)
+						ipoib_put_ah(neigh->ah);
+					ipoib_neigh_free(dev, neigh);
+					continue;
+				}
+			}
+
 			while ((skb = __skb_dequeue(&neigh->queue)))
 				__skb_queue_tail(&skqueue, skb);
 		}
@@ -520,7 +544,25 @@ static void neigh_add_path(struct sk_buff *skb, struct net_device *dev)
 		memcpy(&neigh->dgid.raw, &path->pathrec.dgid.raw,
 		       sizeof(union ib_gid));
 
-		ipoib_send(dev, skb, path->ah, IPOIB_QPN(skb->dst->neighbour->ha));
+		if (ipoib_cm_enabled(dev, neigh->neighbour)) {
+			if (!ipoib_cm_get(neigh))
+				ipoib_cm_set(neigh, ipoib_cm_create_tx(dev, path, neigh));
+			if (!ipoib_cm_get(neigh)) {
+				list_del(&neigh->list);
+				if (neigh->ah)
+					ipoib_put_ah(neigh->ah);
+				ipoib_neigh_free(dev, neigh);
+				goto err_drop;
+			}
+			if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE)
+				__skb_queue_tail(&neigh->queue, skb);
+			else {
+				ipoib_warn(priv, "queue length limit %d. Packet drop.\n",
+					   skb_queue_len(&neigh->queue));
+				goto err_drop;
+			}
+		} else
+			ipoib_send(dev, skb, path->ah, IPOIB_QPN(skb->dst->neighbour->ha));
 	} else {
 		neigh->ah  = NULL;
 
@@ -538,6 +580,7 @@ err_list:
 
 err_path:
 	ipoib_neigh_free(dev, neigh);
+err_drop:
 	++priv->stats.tx_dropped;
 	dev_kfree_skb_any(skb);
 
@@ -640,7 +683,12 @@ static int ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev)
 
 		neigh = *to_ipoib_neigh(skb->dst->neighbour);
 
-		if (likely(neigh->ah)) {
+		if (ipoib_cm_get(neigh)) {
+			if (ipoib_cm_up(neigh)) {
+				ipoib_cm_send(dev, skb, ipoib_cm_get(neigh));
+				goto out;
+			}
+		} else if (neigh->ah) {
 			if (unlikely(memcmp(&neigh->dgid.raw,
 					    skb->dst->neighbour->ha + 4,
 					    sizeof(union ib_gid)))) {
@@ -805,6 +853,7 @@ struct ipoib_neigh *ipoib_neigh_alloc(struct neighbour *neighbour)
 	neigh->neighbour = neighbour;
 	*to_ipoib_neigh(neighbour) = neigh;
 	skb_queue_head_init(&neigh->queue);
+	ipoib_cm_set(neigh, NULL);
 
 	return neigh;
 }
@@ -818,6 +867,8 @@ void ipoib_neigh_free(struct net_device *dev, struct ipoib_neigh *neigh)
 		++priv->stats.tx_dropped;
 		dev_kfree_skb_any(skb);
 	}
+	if (ipoib_cm_get(neigh))
+		ipoib_cm_destroy_tx(ipoib_cm_get(neigh));
 	kfree(neigh);
 }
 
@@ -1080,6 +1131,8 @@ static struct net_device *ipoib_add_port(const char *format,
 
 	ipoib_create_debug_files(priv->dev);
 
+	if (ipoib_cm_add_mode_attr(priv->dev))
+		goto sysfs_failed;
 	if (ipoib_add_pkey_attr(priv->dev))
 		goto sysfs_failed;
 	if (device_create_file(&priv->dev->dev, &dev_attr_create_child))