[ofa-general] [PATCH] ipoib: defer skb_orphan() until irqs enabled

[Roland Dreier]

 > I don't think there is a problem. The only SKB which is subject to
 > this race, is the one we that we posted right after stopping the net
 > queue. But the interrupt handler (resulting from arming the CQ) and
 > possibly the following timer invocations, will drain the CQ up to the
 > point where there are half the queue outstanding WRs. But and this one
 > is at the other half of the queue.

Maybe I'm missing something but where is the logic that stops draining
the CQ?  I just see

static int poll_tx(struct ipoib_dev_priv *priv)
{
	int n, i;

	n = ib_poll_cq(priv->send_cq, MAX_SEND_CQE, priv->send_wc);
	for (i = 0; i < n; ++i)
		ipoib_ib_handle_tx_wc(priv->dev, priv->send_wc + i);

	return n == MAX_SEND_CQE;
}

and

static void drain_tx_cq(struct net_device *dev)
{
	struct ipoib_dev_priv *priv = netdev_priv(dev);
	unsigned long flags;

	spin_lock_irqsave(&priv->tx_lock, flags);
	while (poll_tx(priv))
		; /* nothing */

which seem like they could easily poll that last completion.