[OpenAFS] OpenAFS 1.4.x on SuSE 10.1 and rlim error

Marcus Watts mdw@umich.edu
Fri, 22 Sep 2006 01:51:23 -0400

Jeffrey Hutzelman <jhutz@cmu.edu> writes:
> Date: Wed, 20 Sep 2006 21:09:02 -0400
> From: Jeffrey Hutzelman <jhutz@cmu.edu>
> To: Marcus Watts <mdw@umich.edu>,
>         "OpenAFS-info (E-mail)" <openafs-info@openafs.org>
> cc: Jeffrey Hutzelman <jhutz@cmu.edu>
> Subject: Re: [OpenAFS] OpenAFS 1.4.x on SuSE 10.1 and rlim error 
> On Wednesday, September 20, 2006 05:04:39 PM -0400 Marcus Watts 
> <mdw@umich.edu> wrote:
> > Jeffrey Hutzelman <jhutz@cmu.edu> wrote:
> > ...
> >> ... but often, the problem is that the configure tests utterly failed
> >> because configure was unable to install your kernel headers, perhaps
> >> because you haven't actually installed them.
> > ...
> >
> > Having been just burned by variations on this exact problem,
> > I've now got:
> > 	/afs/umich.edu/user/m/d/mdw/build/tmp/,m20a
> > a diff which will fix the 2.6 kernel configuration process to:
> > /1/ log more useful information in config.log when the
> > 	kernel configuration process breaks -- the error message,
> > 	make command, the compiler lines, the module makefile, and
> > 	the c source for the test.
> > /2/ does a dry run of the kernel build process and errors out with
> > 	a more helpful message if this fails.  If this breaks,
> > 	there is no point in doing further kernel configuration tests,
> > 	and better to tell the user what to do next.
> >
> > This will be part of the rxk5 changes I'm working on, and I hope to
> > have even more improvements to work better with exotic kernel
> > configurations and uml.  But I believe this diff is probably useful
> > as is right now.
> With the caveat that I haven't actually reviewed this yet, I'd suggest
> submitting it to openafs-bugs separately, so that it can be integrated
> and make its way to stable releases more quickly than rxk5 probably will.

Those were "really fresh" diffs -- I hadn't done much testing either.
Since I wrote them in response to seeing the identical kinds of error
messages that Dave Lewis had reported, I figured he might actually
find this of immediate and direct interest.

I'll submit a very slight variation of this to info-bugs@ - basically,
with a comment in the front and a more permanent home.  This is
the "conservative" patch, so sure, getting this into stable makes a lot
of sense.

I also will have a "radical" patch that goes a lot farther - it
detects more problems more directly (for instance, being given
--with-afs-sysname=i386_linux24 but handed a 2.6 kernel), and
it will also go dig more kernel configuration information out
of the kernel build rather than making local guesses.  This makes
the uml build a lot cleaner (which is my immediate desire), but
it should also simplify support for other 'exotic' configurations, such
odd i386 configurations, also amd64, parisc, s390x, sparc64, ppc, etc.
That is, if I haven't instead actually just broken things there.
That is, if it wasn't broken or obselete already.  This patch may
be too radical for the 'stable' track, but there may be aspects
that would be useful to backport.

				-Marcus Watts