[openib-general] user_mad questions

[Sean Hefty]

I ran into a couple of questions while updating the user_mad code that I'd like
to clarify.

First, the only failed sends that are reported are response timeouts.  Is there
a reason that other failures are not reported?  (I can understand treating the
MADs as unreliable, but just wanted to check that this was the intent.)

Second, when a failed send is reported, there's a mismatch between the size of
the data copied to the user and the size returned from ib_umad_read().  The
copied data is sizeof(ib_user_mad) + sizeof(ib_mad).  The reported size turns
out to be sizeof(ib_user_mad) + sizeof(ib_mad_hdr).  It appears that the data
copy is off.

Finally, when reporting that the buffer for a received MAD is too small, what
information actually needs to make it back to the user?  The code currently
copies the entire first segment.  What would happen if only the common, RMPP,
and class specific headers were returned to the user?

- Sean