[ofa-general] cmatose fails whereas rping passes on iWarp

[Roland Dreier]

 > At the same time, I don't want to turn it into if an iWarp device
 > does not do what an Infiniband device would do then it is a bug.
 > I didn't take this issue that way as everyone said it with a :-)
 > I'm probably making it a bigger issue than it is so I will stop now.

I don't think we necessarily choose IB or iWARP semantics.  In this case
for historical reasons (IB was implemented first) we have an interface
with two ways of saying whether sends should be signaled -- the iWARP
way of having only one place to set this is probably cleaner but it's
not a big enough deal to break the interface.  In other places we've
used iWARP semantics -- eg one could view the RDMA CM stuff as doing a
*lot* of work to provide iWARP "connect by IP" semantics on top of IB.

Basically we're trying to come up with a common set of "RDMA verbs" that
can be used on both IB and iWARP hardware.

 - R.