Re: [microblaze-uclinux] Loadable module support for Petalinux w/MMU?

[Steven Hein <ssh@xxxxxxx>]

Hi John,

You were exactly right....the R_MICROBLAZE_32 case
was causing the string problem.   I changed the code as you
suggested, and the module works correctly when compiled
with the normal (glibc) compiler.     (I was thinking that I
had tried to change this earlier, but I think I had ended up
removing the "*location =" statement altogether...which
obviously didn't work!).

Given that this change makes the R_MICROBLAZE_32
and R_MICROBLAZE_64 reloc cases consistent, it
seems that it would be good to get the R_MICROBLAZE_32
change into the SVN trunk so that both are consistent:

               case R_MICROBLAZE_32:
#if __GNUC__ < 4
                       old_value = *location;
                       *location = value + old_value;
#else
                       *location = value;
#endif
                       DBPRINTK("R_MICROBLAZE_32 (%08lx->%08lx)\n",
                               old_value, value);


Thanks for help!      (As I reviewed all of the emails on this,
it seems that all of this has been stated before....so I apologize
for the rehash!     Hopefully summarizing both the 32 and
64-bit reloc issues together helps some other folks......).

Steve



John Williams wrote:
> Hi Steve,
>
> Thanks for your detailed notes, please see my comments below:
>
> Steven Hein wrote:
>
> > * Using petalinux-mmu-v0.10 without any mods, the printk's
> > in my loadable modules failed, as previously documented.
> >
> > * Using petalinux-mmu-v0.10 with John's fix (to #ifdef out
> > the split relocs fix), the printk's worked, but the strings
> > initialized within a structure (i.e. in my test code below) failed.
> >
> > * Using the latest SVN kernel code (currently rev. 4469), which
> > contains the split relocs fix, it failed as above (printk's worked,
> > strings in structures were garbled).
>
> One more thing to try, in the R_MICROBLAZE_32 reloc we also have the 
> "old_value=...; value+=" construct.
>
> Please try changing that to just "*location = value;" and see if it 
> fixes the strings issue.
>
> > * When I remove the split relocs fix, and build the kernel module
> > with the uclinux compiler (microblaze-uclinux-gcc), it works
> > correctly (printk's and strings work) when loaded into the
> > MMU kernel (glibc built).
> >
> > So I have a solution that works: build the modules with
> > the microblaze-uclinux-gcc compiler.     Is there a better solution
> > at this point?    Am I doing something to cause this?
>
> It's certainly not your fault, it's an interaction between the 
> toolchain/linker and the module loader.
>
>
> Cheers,
>
> John
> ___________________________
> microblaze-uclinux mailing list
> microblaze-uclinux@xxxxxxxxxxxxxx
> Project Home Page : http://www.itee.uq.edu.au/~jwilliams/mblaze-uclinux
> Mailing List Archive : 
> http://www.itee.uq.edu.au/~listarch/microblaze-uclinux/

___________________________
microblaze-uclinux mailing list
microblaze-uclinux@xxxxxxxxxxxxxx
Project Home Page : http://www.itee.uq.edu.au/~jwilliams/mblaze-uclinux
Mailing List Archive : http://www.itee.uq.edu.au/~listarch/microblaze-uclinux/