[openib-general] Re: CMA disconnect

[Or Gerlitz]

On 5/8/06, Sean Hefty <mshefty at ichips.intel.com> wrote:
> Or Gerlitz wrote:
> > Looking in the code i have realized that it is a must for the CMA
> > consumer to call rdma_disconnect to have the QP state moved into ERROR.

> Maybe it would make sense for the CMA to transition the QP to the error state
> before destroying it?

It makes sense to it do as a strict/cleanup policy, but **only** if
the user did not cause the CMA to move the QP to ERROR **before** by
calling rdma_disconnect. For example in iSER code we wait to get both
the DISCONNECTED event and completions/FLUSHES on all the posted RX/TX
before calling rdma_destroy_id, so deffering the QP state change to
ERROR will deadlock our design.

Or.