[ofw] [RFC] [PATCH] winverbs: changes to support rdma_cm

[Fab Tillier]

>>> @@ -1202,8 +1236,7 @@
>>>                 ) PURE;
>>>
>>>         STDMETHOD(Disconnect)(
>>> -               THIS_
>>> -               __in_opt OVERLAPPED* pOverlapped
>>> +               THIS
>>
>> Is Disconnect just an EP function, or does it also change the
>> underlying QP's state?  If the latter having it be an overlapped
>> would be helpful.  If just a CM function, the QP Modify is still an
>> overlapped (for flushing)?
>
> Currently, disconnect is purely a CM function.  The QP modify call was
> unchanged and has an overlapped structure.
>
> I wasn't sure that disconnect should modify the QP.  If it doesn't
> then an app can use disconnect to detach the send queue only, and let
> disconnect from the peer detach the receive queue.  (Obviously, it's
> up to the app to enforce this by not sending...)

I think keeping the CM calls separate from the QP calls is probably good.  Not sure how this quite plays out with iWARP close semantics though.  I suppose Disconnect could map to a half-closed TCP connection.

-Fab