[ewg] [PATCH] IB/qib: latest QIB driver fixes

[Roland Dreier]

 > Or - the agreement in Sonoma was that anything submitted to OFED should
 > also be in the process of going into the kernel. Because Roland can't
 > always respond in the time frame required to match the OFED schedule, we
 > explicitly said it didn't have to be accepted. If it turns out that
 > there is some major issue with the change in question ( which has been
 > submitted upstream), we'll respond to that. But I also don't want to
 > either hold up the release, or redo an entire set of patches because of
 > this one (relatively small) change.

<rant>
This is to a large degree hogwash.  The entire qib driver was shipped in
an OFED release long before it was even submitted upstream for the first
time.  So this change in particular is not a big deal but the
development methodology of completely throwing away the upstream ipath
driver and shipping a new codebase (of 50,000+ lines of code!) as part
of OFED seems pretty broken to me.

To be honest I think the qib driver is a symptom of a pretty big
problem.  Looking at the git logs, I see that we merged the ipath driver
around March 2006, and the last patch from qlogic in December 2008.  So
it took you guys about 2.5 years to decide that your first driver was an
unmaintainable mess and had to be thrown away.

I see many signs of the same philosophy of "our needs don't match the
existing kernel facilities perfectly, so let's hack around it or
reimplement it in device-specific driver in an ugly way rather than
working outside our driver to fix the interface" in qib.  Given the
track record of quickly giving up on code as unmaintainable, I'm hoping
we can at least get the qib driver into shape where the community can
continue to keep things together if you guys give up on it.
</rant>

 - R.