Re: [PATCH] modpost: allow modpost to fail on warnings

[Masahiro Yamada]

On Mon, Sep 21, 2020 at 11:51 PM Pierre-Louis Bossart
<pierre-louis.bossart@xxxxxxxxxxxxxxx> wrote:
>
> Thanks for the review,
>
> >> Set KBUILD_MODPOST_FAIL_ON_WARNINGS to a non-empty value to make the
> >> kbuild fail when modpost generates any warnings. This will avoid
> >> misses such as [1] where the SOF CI did not catch a missing module
> >> license.
> >>
> >> This was initially contributed in 2016 [2], rebase/clean-ups and tests
> >> by Pierre Bossart.
> >>
> >> Test example:
> >> $ KBUILD_MODPOST_FAIL_ON_WARNINGS=1 make
> >>    GEN     Makefile
> >>    DESCEND  objtool
> >>    CALL    sof-dev/scripts/atomic/check-atomics.sh
> >>    CALL    sof-dev/scripts/checksyscalls.sh
> >>    CHK     include/generated/compile.h
> >>    MODPOST Module.symvers
> >> Kernel: arch/x86/boot/bzImage is ready  (#13)
> >> WARNING: modpost: missing MODULE_LICENSE() in sound/soc/intel/boards/snd-soc-sof-sdw.o
> >> make[2]: *** [sof-dev/scripts/Makefile.modpost:114: Module.symvers] Error 2
> >
> >
> > I think [1] should be an error instead of a warning
> > by default.
>
> would the following patch be what you have in mind?


No.
error() does not exist.

merror() exists, but the difference from warn()
is just a prefix.

If any error happens, modpost should return the error code.





> diff --git a/scripts/mod/modpost.c b/scripts/mod/modpost.c
> index 422f1cfca289..ae1eb67aa0f2 100644
> --- a/scripts/mod/modpost.c
> +++ b/scripts/mod/modpost.c
> @@ -2018,7 +2018,7 @@ static void read_symbols(const char *modname)
>          if (!mod->is_vmlinux) {
>                  license = get_modinfo(&info, "license");
>                  if (!license)
> -                       warn("missing MODULE_LICENSE() in %s\n", modname);
> +                       error("missing MODULE_LICENSE() in %s\n", modname);
>                  while (license) {
>                          if (license_is_gpl_compatible(license))
>                                  mod->gpl_compatible = 1;
>
>
> If yes, also wondering if we can still add the option to treat warnings
> as errors as an opt-in behavior?


I want to add a new option only when it is necessary to do so.

I am not sure which warnings are real warnings.


> Thanks!
> -Pierre



-- 
Best Regards
Masahiro Yamada