Bug#701690: nvidia: disagrees about version of symbol efi_enabled

[Ben Hutchings]

On Tue, Feb 26, 2013 at 10:54:42PM +0100, Andreas Beckmann wrote:
> On 2013-02-26 14:44, Ben Hutchings wrote:
> > On Tue, 2013-02-26 at 13:24 +0000, Ben Hutchings wrote:
> >> On Tue, 2013-02-26 at 10:35 +0200, Antti Salmela wrote:
> >>> ABI seems to have changed with the latest 3.2.39-1 kernel. Proprietary
> >>> nvidia driver (313.18-2)  from experimental fails to load if compiled
> >>> against previous (or maybe even original...) linux-headers-3.2.0-4-amd64
> >>> packages, but works after recompile.
> 
> Antti, did you use the -dkms package or the -source package and
> module-assistant?
> 
> >>> [   14.383562] nvidia: module license 'NVIDIA' taints kernel.
> >>> [   14.386689] nvidia: disagrees about version of symbol efi_enabled
> >>> [   14.386692] nvidia: Unknown symbol efi_enabled (err -22)
> >> [...]
> >>
> >> Sorry, I ignored this ABI change because codesearch.debian.net didn't
> >> show the symbol being used by out-of-tree modules.  Perhaps it doesn't
> >> include non-free code, or this was changed between unstable and
> >> experimental.
> > 
> > Oh, now I see, the code is packaged up in a way that can't be searched.
> > 
> > Ben.
> > 
> >> I'll see what I can do to fix this.  (And unfortunately if the change is
> >> reverted then you'll need to rebuild the module again.)
> 
> Ben, Should I upload nvidia-graphics-modules rebuilt against 3.2.39-1?
 
No, I believe I can fix this.

Also, if you rebuild without changing the nvidia source you'll
introduce a bug, because the test of efi_enabled will always return
true now that it's the name of a function...

Ben.

-- 
Ben Hutchings
We get into the habit of living before acquiring the habit of thinking.
                                                              - Albert Camus