[lug] pte_offset in a device driver isn't being found

[Michael J. Hammel]

On Sat, 2003-06-21 at 12:00, Jonathan Corbet wrote:
> Good thinking - seeing as it's a macro, things will work poorly if you
> don't have the definition around.  You'll not get your module to load.

Exactly.  I dug through the headers but can't find a "pte_offset()"
macro - only a "__pte_offset()" and some pte_offset_xxx(), where xxx is
"kernel", "map" or "nested".  pte_offset_kernel() seemed to match the
args so I switched to that, but I doubt that's correct.

> The macro, for the x86 architecture (is that what you're on?) 

Yes, asm-i386.

> is defined in
> asm-i386/pgtable.h.  It is inside an ifdef:
> 
> 	#ifndef __ASSEMBLY__
> 
> 	#endif
> 
> But that *shouldn't* be a problem.  Something to check, though.

I don't see any references to ASSEMBLY in the driver code.

> > I'm including /usr/include and /usr/src/linux-<uname -r>/include in the
> > header search path.
> 
> For kernel builds, you really shouldn't be getting anything from
> /usr/include at all.  Take it out of the path, it can only cause trouble.

I added it myself as a test, but the original Makefile used the kernel
dirs only.  I can take it out but it doesn't change things.  The problem
is that the pte_offset() macro doesn't appear to exist in the 2.4.20
kernel headers for i386.  It *does* exist in the headers for other
platforms - I grep'd them and found it in other pgtable.h files for
those platforms.  It's just not there in i386.

-- 
Michael J. Hammel                               The Graphics Muse 
mjhammel at graphics-muse.org                      http://www.graphics-muse.com
------------------------------------------------------------------------------
    Great spirits have always encountered violent opposition from mediocre
                         minds.  --  Albert Einstein