> And it still doesn't. > Jack here also confirms that the problem still exits if SM clears P_Key table > and then later readds a P_Key. > > Could you take a look please? So is the solution to call ipoib_pkey_dev_delay_open() in ipoib_ib_dev_flush()?