Re: [GIT PULL] Kbuild updates for v6.15-rc1

[Stephen Rothwell]

Hi Linus,

On Sun, 6 Apr 2025 09:50:05 -0700 Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Sat, 5 Apr 2025 at 22:43, Damian Tometzki <damian@xxxxxxxxxxxxxx> wrote:
> >
> > i got the following error after this pull request.
> >
> >  MODPOST Module.symvers
> > ERROR: modpost: missing MODULE_DESCRIPTION() in lib/tests/slub_kunit.o
> > make[3]: *** [/home/damian/kernel/linux/scripts/Makefile.modpost:147: Module.symvers] Error 1
> > make[2]: *** [/home/damian/kernel/linux/Makefile:1956: modpost] Error 2
> > make[1]: *** [/home/damian/kernel/linux/Makefile:248: __sub-make] Error 2
> > make[1]: Leaving directory '/home/damian/kernel/build'
> > make: *** [Makefile:248: __sub-make] Error 2  
> 
> Bah. And the reason it doesn't show up in my normal build tests is
> that my 'allmodconfig' tests end up picking up SLUB_TINY - which
> disables a lot of slub cases, including this test.
> 
> And my normal non-allmodconfig tests don't enable SLUB_KUNIT_TEST.
> 
> That said, I'm not sure if making it a hard error was a good idea in
> the first place. It *used* to be just a warning, and it used to be
> enabled only with 'extra_warn'.
> 
> So switching it on to always warn was probably a good idea, but then
> also making it a hard error may have been a bit excessive.
> 
> In related news - I also wonder whether SLUB_TINY should just be
> turned off for COMPILE_TEST. It's not a very interesting config option
> to test for, and it disables much more code than it enables [*].
> 
> Testing this without SLUB_TINY, I see that it *also* triggers this one:
> 
>     ERROR: modpost: missing MODULE_DESCRIPTION() in mm/kasan/kasan_test.o
> 
> so the claim in commit 6c6c1fc09de3 ("modpost: require a
> MODULE_DESCRIPTION()") that "all known instances of this issue have
> been fixed" is clearly wrong.
> 
> For all we know, there are lots of other cases like this that just
> never got tested with W=1 at all.
> 
> I think I'll downgrade the error() to a warn() again, and make
> SLUB_TINY depend on !COMPILE_TEST.
> 
> And I'm not even convinced we should require module descriptions for
> silly test modules, but whatever.
> 
> We'll see if something else pops up, but making the lack of a module
> description a fatal error was clearly not right as-is.

https://lore.kernel.org/lkml/20250324103048.3d8230f9@xxxxxxxxxxxxxxxx/

-- 
Cheers,
Stephen Rothwell

Attachment: pgp1J0kTAI7vY.pgp
Description: OpenPGP digital signature