[openib-general] RDMA CM and loopback addresses

[Sean Hefty]

>Hmm. This all sounds very reasonable, but if resolve_addr is
>available, why not get rid of bind()? Do not provide any way
>in the API for a non-listening socket to get the bound hardware device
>without providing both source and dest.

This makes sense, and was the way the initial implementation was coded.

Support for bind() on the active side was added to support DAPL.  In fact,
calling bind() doesn't even alleviate the need to call resolve_addr().

I will see if there's find a reasonable way to support DAPL without binding
early to a device.

- Sean