Re: git: f7026fbbb2bd - main - sys/modules: MODULES_OVERRIDE takes precedence over EXTRA_MODULES and WITHOUT_MODULES

On Thu, Jan 5, 2023 at 3:51 AM Andriy Gapon <andriy.gapon@uabsd.com> wrote:

> On 2022-10-29 19:33, Warner Losh wrote:
> > The branch main has been updated by imp:
> >
> > URL:
> https://cgit.FreeBSD.org/src/commit/?id=f7026fbbb2bd72176d73921dd2dd5e4b263d5103
> >
> > commit f7026fbbb2bd72176d73921dd2dd5e4b263d5103
> > Author:     Warner Losh <imp@FreeBSD.org>
> > AuthorDate: 2022-10-29 14:34:16 +0000
> > Commit:     Warner Losh <imp@FreeBSD.org>
> > CommitDate: 2022-10-29 16:17:40 +0000
> >
> >      sys/modules: MODULES_OVERRIDE takes precedence over EXTRA_MODULES
> and WITHOUT_MODULES
> >
> >      MODULES_OVERRIDE has traditionally taken precedence over
> EXTRA_MODULES
> >      and WITHOUT_MODULES as the exact list of modules to build. Over
> time,
> >      things have been added that has broken this. Move the .endif that
> makes
> >      this the case to the right place. The so called 'ALL_MODULES'
> option is
> >      the only thing with higher precedence, but it's not quite all the
> >      options anymore (though it is much more of them, and doesn't quite
> >      work on !x86).
> >
> >      Sponsored by:           Netflix
>
> Warner,
>
> this change broke one use-case of mine.  The use-case does not seem to
> be exotic, so others may be affected as well.
>
> I am building a custom arm64 kernel and a set of modules for a specific
> SoC (rockchip).
> The kernel configuration, similarly to GENERIC, has this statement:
>         include         "std.rockchip"
>
> In turn, std.rockchip has this statement:
>         # DTBs
>         makeoptions     MODULES_EXTRA+="dtb/rockchip"
>
> Additionally, in my make.conf I have MODULES_OVERRIDE set to a short
> list of modules.
>
> Previously this worked without issues, completely transparently.
> Now, MODULES_OVERRIDE apparently cancels MODULES_EXTRA.
> I can easily add dtb/rockchip to my MODULES_OVERRIDE, of course, but I
> get a feeling that the previous behavior was more useful because it
> allowed to specify mandatory "internal" modules in addition to a
> user-specified list of modules.  Not sure what's the benefit of the
> traditional behavior that you restored, it's not spelled out.
>
> TLDR: dtb-s and dtbo-s are not getting build for anyone with
> MODULES_OVERRIDE in make.conf (on relevant platforms).
>

I'm swamped after the holidays. I've put this in my queue to consider, but
it may be a week or three before I get to it. Please hassle me in a couple
weeks if I somehow drop this on the floor. It seems like there's an issue
here, but I need to look through the details and I won't have the time for
a few days...

Warner


> > ---
> >   sys/modules/Makefile | 4 ++--
> >   1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/sys/modules/Makefile b/sys/modules/Makefile
> > index 00afbffb1baf..3a009f071cab 100644
> > --- a/sys/modules/Makefile
> > +++ b/sys/modules/Makefile
> > @@ -848,8 +848,6 @@ _nvram=           opal_nvram
> >   _nvram+=    powermac_nvram
> >   .endif
> >
> > -.endif
> > -
> >   .if ${MACHINE_CPUARCH} == "arm" || ${MACHINE_CPUARCH} == "aarch64"
> >   _bcm283x_clkman=  bcm283x_clkman
> >   _bcm283x_pwm=  bcm283x_pwm
> > @@ -867,6 +865,8 @@ SUBDIR+=${MODULES_EXTRA}
> >   SUBDIR:= ${SUBDIR:N${reject}}
> >   .endfor
> >
> > +.endif # MODULES_OVERRIDE -- Keep last
> > +
> >   # Calling kldxref(8) for each module is expensive.
> >   .if !defined(NO_XREF)
> >   .MAKEFLAGS+=        -DNO_XREF
>
> --
> Andriy Gapon
>
> https://standforukraine.com
> https://razomforukraine.org
>
>