On 01/19/2016 12:04 PM, Moni Shoua wrote: > On Mon, Jan 18, 2016 at 11:22 PM, Doug Ledford wrote: >> rdmavt >> qib->rdmavt changes >> >> Things that I intended to submit after the merge window opened and >> Greg's merge was done: >> >> Retake staging/rdma >> Delete staging/rdma/{amso1100,ehca,ipath} >> hfi1->rdmavt changes >> Add staging/rdma/soft-rxe >> > Doug, > it's not clear if you intend to take rdmavt into drivers/infiniband/sw I had intended to. However, on Jan 6th at roughly 1pm EST, Dennis submitted v3 of his rdmavt patch set. At roughly 2pm EST you submitted a completely different, competing patchset that resides in the same place in the filesystem, has the same name, but is 100% different internally. I know Intel and Mellanox have a hard time working together, but geeze... > IMHO such a central piece of code needs to go through proper design > stages, be written with a clear and sharp interface and be published > with documentation. Yes, you've repeated this multiple times. As this is a kernel internal interface only, and a private one to the IB stack, we can be a bit more flexible than you suggest here and do things a little more on the fly. > To me it looks like the only purpose of rdmavt is to pull hfi1 driver > out from staging. That's obviously the piece that Intel cares about, sure. > If we want it as an infrastructure for any kind of > low level driver it needs to stick to infrastructure standards. This statement is too vague for me to say anything constructive. > I hope that you agree Again, too vague to answer. -- Doug Ledford GPG KeyID: 0E572FDD