From 839fcaba355abaffb7b44f0f4504093acb0b11cf Mon Sep 17 00:00:00 2001 From: "Michael S. Tsirkin" Date: Mon, 5 Feb 2007 22:12:23 +0200 Subject: 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 Signed-off-by: Roland Dreier --- drivers/infiniband/ulp/ipoib/ipoib_main.c | 63 ++++++++++++++++++++++++++++--- 1 file changed, 58 insertions(+), 5 deletions(-) (limited to 'drivers/infiniband/ulp/ipoib/ipoib_main.c') diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c index af5ee2ec449..18d27fd352a 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_main.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c @@ -49,8 +49,6 @@ #include -#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)) -- cgit v1.2.3