[openib-general] Re: ipoib and core patches for 2.6.17

[Roland Dreier]

    Michael> Guys, I'm worried that several critical IMO issues
    Michael> reported (with patches) a long time ago still aren't
    Michael> fixed even on svn trunk.  I agree the issues are
    Michael> non-trivial races, but I wander whether you could give
    Michael> these a bit higher priority?

    Michael> Roland, could we get the discussion going again on my
    Michael> pile of ipoib patches?

Yes, I haven't been as fast as I should have been.  I think the
current status is the following:

    ipoib_all_neigh_issues_2.patch:
        There's a better fix pending with Dave Miller.
    ipoib_flush_wq_1.patch:
        I need to dedicate some time to looking at this.  I don't like
        putting more stuff into the ipoib workqueue, so I want to look
        for a different solution to the fix.
    ipoib_flush_wq_2.patch:
        Looks fine.  Does this depend on ipoib_flush_wq_1.patch?
    ipoib_init_qp.patch:
        I don't see why it's worth applying this.  It just seems like
        churn that doesn't necessarily improve anything.
    ipoib_multicast_race.patch:
        Looks fine.  I'll apply now.
    ipoib_post_receives_err.patch:
        Looks fine, I guess.  I'll apply now.
    ipoib_qprst_protect.patch:
        This looks like it clutters the code with locking even though
        the race just causes one extra trip through the loop.  So I
        don't see why we care about it.

So the only patch where I'm being a problem is ipoib_flush_wq_1.patch.

 - R.