[openib-general] librdmacm code confusion wrt iWarp

[Sean Hefty]

> I think the kernel could probably also be common and it would work.
> iWARP QPs don't have a separate RESET state like IB, so the qp_modify
> RESET->INIT really becomes a noop.  iWARP QPs start out in IDLE state
> and its valid to post receives in IDLE.  We map iWARP IDLE -> IB INIT
> and IB RESET.

Part of the call to transition the IB QP to INIT involves looking up the 
pkey_index.  Is this simulated as well?

My intention is to change the librdamcm to call rdma_init_qp_attr() when 
transitioning the IB QP from reset->init.  (This is necessary to get the qkey 
used by the ipoib port space.)  So, the userspace code should get the same 
qp_attr and qp_attr_mask as that used in the kernel when a QP is created.

- Sean