[ofw] RE: Release IBA resources in DISPATCH_LEVEL ?

[Sean Hefty]

>Don't you have to have the logic today to delay requests and issue them once a
>prerequisite request completes?  Your example of the destroy CQ command having
>to be delayed if a destroy QP associated with that CQ is in flight seems like
>an example of a case where no matter what your software usage mode is, you have
>to delay issuing the destroy CQ until the destroy QP completes.

If destruction were asynchronous, then in this example the driver/verbs layer
should just return an error.  Destroy CQ should be deferred until the QP
destruction completes, versus being initiated.

I don't know how much async destruction helps.  At some point, the ULP needs to
wait for the async destruction to complete, and I don't know why the ULP
couldn't just do sync destruction at that point.  Async operations for other
memory registration or QP modification seems beneficial however.

- Sean