[ofa-general] IPoIB kernel Oops -- race condition

[Jack Morgenstein]

On Monday 29 June 2009 07:14, Jack Morgenstein wrote:
> > 
> On second thought, maybe it would be simpler to just create an ipoib_stop_task(),
> and do everything ipoib_stop() does in that workqueue task. leave would thus always
> be executed in the workqueue.
> 
> Thoughts?
> 
Looked at this again.  There is a problem with this approach -- If I do ipoib_stop in
the workqueue, I need to do ipoib_open in the workqueue as well.  In ipoib_open, I will
lose all the error returns if I take this approach:

int ipoib_open(struct net_device *dev)
{
	struct ipoib_dev_priv *priv = netdev_priv(dev);

	ipoib_dbg(priv, "bringing up interface\n");

	napi_enable(&priv->napi);
	set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);

	if (ipoib_pkey_dev_delay_open(dev))
		return 0;

	if (ipoib_ib_dev_open(dev)) {
		napi_disable(&priv->napi);
		return -EINVAL;
	}

	if (ipoib_ib_dev_up(dev)) {
		ipoib_ib_dev_stop(dev, 1);
		napi_disable(&priv->napi);
		return -EINVAL;
	}

	if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
		struct ipoib_dev_priv *cpriv;

		/* Bring up any child interfaces too */
		mutex_lock(&priv->vlan_mutex);
		list_for_each_entry(cpriv, &priv->child_intfs, list) {
			int flags;

			flags = cpriv->dev->flags;
			if (flags & IFF_UP)
				continue;

			dev_change_flags(cpriv->dev, flags | IFF_UP);
		}
		mutex_unlock(&priv->vlan_mutex);
	}

	netif_start_queue(dev);

	return 0;
}

I'm very worried that if we do things asymmetrically we will introduce instability.
Regarding doing only the multicast leave/free in the workqueue -- I do not see how to do
this cleanly (i.e., everything as is, but simply scheduling a leave-free task).

Yossi, Roland, Eli?

-Jack