[ofa-general] Re: [PATCH] IB/srp: enforce protocol limit on srp_sg_tablesize

[David Dillow]

On Fri, 2008-03-21 at 19:51 -0700, Roland Dreier wrote:
> > I see you queued this for 2.6.26 -- I'd have thought that this would go
>  > to 2.6.25 as it is a bug fix. What do you think?
> 
> I didn't think it was serious enough to go in at the rc5/rc6 stage.
> It's borderline even to call it a bug, and the workaround is pretty
> trivial -- "don't do that then."

While certainly we'd want to avoid large changes for minor bugs this
late in the -rc cycle, a small, obvious change to ensure protocol
correctness would be fine in my opinion. The problem with the workaround
is that you won't hit the exception case until you have memory
fragmentation, and then you get a dropped/hung SRP connection or in the
hopefully unlikely case, data corruption. And even then you're not
likely to realize that you made the sg table too large -- it does not
have an obvious link to the symptoms.

That said, I'm not vehement about it waiting an extra cycle. I know to
keep my sg table entries down, and hopefully others that don't know
there is an issue will be reading the list. Since it won't be in an OFED
release for a while, I'd bet they're unlikely to be upgrading to 2.6.25
in any event.
-- 
Dave Dillow
National Center for Computational Science
Oak Ridge National Laboratory
(865) 241-6602 office