[OpenAFS-devel] Re: OpenAFS 1.4.7pre1 Oops

[Jeffrey Hutzelman]

--On Saturday, April 05, 2008 09:25:51 PM -0400 Marc Dionne 
<marc.c.dionne@gmail.com> wrote:

> 05/04/08 07:59 PM, Simon Wilkinson:
>> As far as I can tell, the configure test for INIT_MM is bogus on 2.6,
>> which has an empty modversions.h. When I asked about detecting the
>> 'exported' state of a kernel symbol, I was told the only way to do so
>> was via a weak reference at runtime.
>
> Yes the test is bogus. I had a reworked version of the test in the second
> patch (see the v2 patch in the ticket), along with some adjustments in
> acinclude.m4.
>
> It doesn't rely on a weak reference.

This really isn't necessary.  The osi_probe code depends heavily on weak 
references for determining which markers it can use during a scan; one more 
is not going to kill you.  Note that without init_mm, we have no way to 
tell whether a potential syscall table address is in range, which means 
that without it probing won't (currently) work on any platform.  It could 
be made to work on some platforms at the expense of omitting the range 
test, but that can cause problems (there's a reason we added that test) 
and, as previously mentioned, is probably not worth the effort given the 
keyring-based PAG code.

-- Jeff