[openib-general] putting in dead wood for DAPL and similar abomination

[Roland Dreier]

    Christoph> Please don't put in things like the address translation
    Christoph> service or memory windows for DAPL folks.  The IB code
    Christoph> in the kernel already has far too much unused stuff and
    Christoph> adding more will not go past reviews for kernel
    Christoph> inclusions - as will DAPL itself exactly because of
    Christoph> such utter stupidities.  Similar hint to the NFS over
    Christoph> RDMA folks at CITI - if you want your stuff to go in
    Christoph> use the openib helper directly below the transport
    Christoph> switch - differnet RDMA transports are too diverse to
    Christoph> be sanely abstracted out and DAPL does a horrible job
    Christoph> at that.  If we need to consolidate code for differnt
    Christoph> transports we can put it into a library later on.

I agree with this sentiment.  (Notice how I asked if any real
applications are using memory windows?)  I also agree that it makes
sense to build abstractions by looking at multiple real
implementations, rather than trying to design the abstractions in
advance.  We're just now beginning to understand how a clean
InfiniBand stack should look, and I haven't seen any free software for
other RDMA transports.

By the way, at least for the code I wrote, anything that doesn't have
a kernel user yet is there because it is used by a real protocol that
should make it upstream eventually.

 - R.