[openib-general] RFC: mthca: implement timewait by tracking QPNs

[Roland Dreier]

    Roland> My gut reaction is that it seems pretty ugly.

    Michael> Hmm. All of it or just some bits?

Well, the idea of pushing timewait handling down into the low-level
drivers seems strange to me.  I don't think any other stack or any
other OS does anything like this.

    Michael> Could be a library function in core so that ipath etc can
    Michael> reuse it.  But note how there's no dependency between
    Michael> drivers here - no reason to block change in mthca until
    Michael> ipath/ehca implement this functionality, too.

I guess the only thing would be that we should implement this for
mthca to maximize the amount ipath/ehca can reuse when they implement this.

    Michael> Not entirely corect. Please look at 9.7.1 - search for
    Michael> "stale packets":

OK, this is somewhat convincing...

    Michael> I don't see how this limits the rate of QP
    Michael> creation. Could you explain?

Once all QPs are tied up in timewait state, then new QPs can only be
created as old QPs leave timewait.  Probably there are enough QPs and
timewait is short enough that this won't be a problem in practice, but
it's the same idea in theory as a busy server running out of fds
because of sockets in timewait state.

 - R.