Linux kernel compatability

[Garrett Wollman]

In article <20110106194030.GA27507 at famine.west.isilon.com>,
zml at freebsd.org writes: 

>Unlike 4 years ago, we now have staffed IB personnel that will be
>working with the code in the FreeBSD tree. We have every incentive to
>make sure IB stays working in head - it saves us so much time when we
>merge. And, for those who heard of our recent acquisition by EMC, we
>have a nod from our parent company to keep doing what we're doing.
>
>I hope this mitigates most of the non-technical concerns.

I *hope* we are clear that we want Infiniband support in the tree.  I
don't think we agree about whether we want the compatibility layer to
grow into a general "Linux KPI support layer" or simply part of the
Infiniband stack in parallel with the adaptation layers used by lots
of other vendor-supported drivers.  So long as it remains what it
currently is -- just the interfaces you need to make your code work --
then there is less chance of it becoming a maintenance problem in the
future.  On the other hand, we already have this issue with other bits
of Linux kernel code that have been ported, more or less successfully,
and their own (much thinner) compatibility glue layers.  The argument
that having only one such layer is better than having many is not
unreasonable; it's just not clear to me that this is maintainable.

-GAWollman