[ofa-general] enabling invalidate operations in libibverbs API

[Roland Dreier]

 > Aren't these invalidate operations valid only for physical memory
 > regions? If that is the case, and since we can't have physical memory
 > regions in userspace, I don't see why we need this in userspace.

invalidate operations are also valid for memory windows, which we will
probably add support for.  And a userspace application could easily be
talking to a kernel driver on the other end, so send w/invalidate etc
also makes sense in that case.

I did leave out all the fast register through a send queue support, of course.

 > I also don't think IB spec defines RDMA READ with invalidate operations.

Correct, but it is an iWARP operation.

 - R.