Another question.
> @@ -769,32 +768,32 @@ static void ipoib_set_mcast_list(struct
> static void ipoib_neigh_destructor(struct neighbour *n)
> {
> struct ipoib_neigh *neigh;
> - struct ipoib_dev_priv *priv = netdev_priv(n->dev);
> + struct ipoib_dev_priv *priv;
> unsigned long flags;
> struct ipoib_ah *ah = NULL;
>
> - ipoib_dbg(priv,
> - "neigh_destructor for %06x " IPOIB_GID_FMT "\n",
> - IPOIB_QPN(n->ha),
> - IPOIB_GID_RAW_ARG(n->ha + 4));
> -
> - spin_lock_irqsave(&priv->lock, flags);
>
> neigh = *to_ipoib_neigh(n);
> if (neigh) {
> + priv = netdev_priv(neigh->dev);
> + ipoib_dbg(priv,
> + "neigh_destructor for %06x " IPOIB_GID_FMT "\n",
> + IPOIB_QPN(n->ha),
> + IPOIB_GID_RAW_ARG(n->ha + 4));
> +
> + spin_lock_irqsave(&priv->lock, flags);
> if (neigh->ah)
> ah = neigh->ah;
> list_del(&neigh->list);
> ipoib_neigh_free(n->dev, neigh);
> + spin_unlock_irqrestore(&priv->lock, flags);
> }
> -
> - spin_unlock_irqrestore(&priv->lock, flags);
> -
> if (ah)
> ipoib_put_ah(ah);
> }
Using to_ipoib_neigh outside priv->lock looks problematic.
Can you convince me this does not introduce new races?
--
MST
_______________________________________________
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
