You are right. This is the one place where priv->lock is not held.
I'm sending V3.
Roland Dreier wrote:
> @@ -587,8 +589,10 @@ void ipoib_mcast_join_task(struct work_s
> __ipoib_mcast_add(dev, priv->broadcast);
> spin_unlock_irq(&priv->lock);
> }
> + rtnl_unlock();
>
> - if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
> + if (priv->broadcast &&
> + !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
I'm trying to understand this patch. What protects access to
priv->broadcast here and prevents it from becoming NULL right after the
test?
- R.
--
--Yossi
_______________________________________________
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
