[OpenAFS-devel] Problem with current cvs and linux 2.4.28-pre4

Jeffrey Hutzelman jhutz@cmu.edu
Wed, 20 Oct 2004 16:34:38 -0400

On Wednesday, October 20, 2004 10:24:22 -0700 Miles Davis 
<miles@cs.stanford.edu> wrote:

> On Wed, Oct 20, 2004 at 11:53:32AM -0500, Neulinger, Nathan wrote:
>> I think the problem is that configure for me has never been able to
>> determine the export status of the symbol, and now the default is to
>> assume that it isn't exported. (Yes, it's been a while since I updated
>> clients... Looks like that was changed 7 months ago or so.) Is there a
>> preferred way of addressing this at build time other than a quick hack?
> I believe the current answer is to modify the configure script to assume
> yes -- see Jeffrey Hutzelman's post from Monday.

That's the short-term answer for 2.6, where eventually we will have a 
better answer.

On a real, stock 2.4 kernel (not one with bizarre patches from redhat or 
whoeever), built with CONFIG_MODVERSIONS, you should not have this problem. 
If your kernel was not built with CONFIG_MODVERSIONS, then the situation 
will be the same as for 2.6 -- we can't tell whether a symbol is exported, 
and so we assume not.  As Miles surmises, the workaround is also the same 
-- patch the configure script.

We'll eventually have a fix for 2.6, which will allow us to determine at 
runtime which symbols are exported, and react accordingly.  However, the 
features in the kernel module loader required to support this do not exist 
in the 2.4 kernel series.

-- Jeff