[openib-general] [PATCH]rmmod could crash in tx_ring[].skb & rx_ring[].skb.

[Roland Dreier]

    Shirley> I am going through the IPoIB code. I found rmmod could
    Shirley> cause kernel crash. It's easy to fix tx_ring with
    Shirley> tx_lock. rx_ring[].skb has the same problem, inducing
    Shirley> another lock in ipoib_ib_handle_wc() just for rmmod is
    Shirley> too costly. Any other suggestion?

Sorry to be so dense, but could you explain the race you're protecting
against?  The code you're putting the lock around can only run 5
seconds after the interface has been shut down, so we're assuming the
hardware is wedged.  But in any case the net queue is stopped so no
new transmits should be able to be posted.

 - R.