> 10-second port failover test has been running with IPoIB UD ipoibtools
> HA for over 8 hours, and there have been very few slow failovers:
>
> $ grep seconds screenlog.7 | wc -l
> 29705
>
> $ grep seconds screenlog.7 | fgrep -v "over 1." | fgrep -v "over 2."
> Interim result: 45.29 10^6bits/s over 53.21 seconds
> Interim result: 299.37 10^6bits/s over 7.34 seconds
> Interim result: 406.76 10^6bits/s over 5.84 seconds
> Interim result: 614.00 10^6bits/s over 3.91 seconds
> Interim result: 579.55 10^6bits/s over 4.06 seconds
> Interim result: 239.60 10^6bits/s over 10.19 seconds
>
> Scott
So it seems we are timing out the connection instead of getting RARP
from the remote and tearing it down outselves.
I wonder whether the following (untested) patch improves things for you.
--
MST
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c
b/drivers/infiniband/ulp/ipoib/ipoib_main.c
index f2a40ae..a6d0594 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_main.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c
@@ -682,6 +682,25 @@ static int ipoib_start_xmit(struct sk_buff *skb, struct
net_device *dev)
}
neigh = *to_ipoib_neigh(skb->dst->neighbour);
+ if (unlikely(memcmp(&neigh->dgid.raw,
+ skb->dst->neighbour->ha + 4,
+ sizeof(union ib_gid))) &&
+ likely(neigh->ah)) {
+ spin_lock(&priv->lock);
+ /*
+ * It's safe to call ipoib_put_ah() inside
+ * priv->lock here, because we know that
+ * path->ah will always hold one more reference,
+ * so ipoib_put_ah() will never do more than
+ * decrement the ref count.
+ */
+ ipoib_put_ah(neigh->ah);
+ list_del(&neigh->list);
+ ipoib_neigh_free(dev, neigh);
+ spin_unlock(&priv->lock);
+ ipoib_path_lookup(skb, dev);
+ goto out;
+ }
if (ipoib_cm_get(neigh)) {
if (ipoib_cm_up(neigh)) {
@@ -689,25 +708,6 @@ static int ipoib_start_xmit(struct sk_buff *skb, struct
net_device *dev)
goto out;
}
} else if (neigh->ah) {
- if (unlikely(memcmp(&neigh->dgid.raw,
- skb->dst->neighbour->ha + 4,
- sizeof(union ib_gid)))) {
- spin_lock(&priv->lock);
- /*
- * It's safe to call ipoib_put_ah() inside
- * priv->lock here, because we know that
- * path->ah will always hold one more reference,
- * so ipoib_put_ah() will never do more than
- * decrement the ref count.
- */
- ipoib_put_ah(neigh->ah);
- list_del(&neigh->list);
- ipoib_neigh_free(dev, neigh);
- spin_unlock(&priv->lock);
- ipoib_path_lookup(skb, dev);
- goto out;
- }
-
ipoib_send(dev, skb, neigh->ah,
IPOIB_QPN(skb->dst->neighbour->ha));
goto out;
}
_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
