[openib-general] ib_req_notify_cq() kernel verb

[Roland Dreier]

    James> Does the ib_req_notify_cq() kernel verb conform the
    James> definition of the Request Completion Notification in
    James> section 11.4.2.2 of the IBTA spec?

    James> Specifically, is the handler only guaranteed to be called
    James> after the next completion entry is added?

Yes, I think these should be semantics we assume.

    James> I understand that Mellanox hardware can generate
    James> notification if there are already completions on the CQ,
    James> but I want to avoid using any hardware specific features.

I agree - we should not assume the stronger Mellanox semantics.

 - R.