[openib-general] [PATCH] IB/ipoib: NAPI

[Roland Dreier]

    Todd> I agree.  This would also mean the ipoib_warn in
    Todd> ipoib_ib_completion would go away (would be a valid
    Todd> situation).

Which warning?  I don't see anything that would change, and I don't
see any warnings at all in ipoib_ib_completion().

    Todd> I'm going to keep thinking about this, seems like we can't
    Todd> call req_notify until after netif_rx_complete, otherwise we
    Todd> have a different race.  That leads to the req_notify and
    Todd> peek approach.

Yes, that's right.  Doing req_notify before netif_rx_complete risks
triggering the event before netif_rx_complete, which leads to the poll
routine never getting scheduled at all.

 - R.