Re: [PATCH] Makefile: remove stale cc-option checks

[Masahiro Yamada]

On Tue, Aug 17, 2021 at 3:36 AM Nick Desaulniers
<ndesaulniers@xxxxxxxxxx> wrote:
>
> On Fri, Aug 13, 2021 at 6:43 PM Masahiro Yamada <masahiroy@xxxxxxxxxx> wrote:
> >
> > On Wed, Aug 11, 2021 at 5:42 AM Nick Desaulniers
> > <ndesaulniers@xxxxxxxxxx> wrote:
> > >
> > > --param=allow-store-data-races=0 was renamed to --allow-store-data-races
> > > in the GCC 10 release.
> > >
> > > diff --git a/Makefile b/Makefile
> > > index 027fdf2a14fe..3e3fb4affba1 100644
> > > --- a/Makefile
> > > +++ b/Makefile
> > > @@ -844,17 +847,17 @@ KBUILD_RUSTFLAGS += -Copt-level=z
> > >  endif
> > >
> > >  # Tell gcc to never replace conditional load with a non-conditional one
> > > -KBUILD_CFLAGS  += $(call cc-option,--param=allow-store-data-races=0)
> > > +ifdef CONFIG_CC_IS_GCC
> >
> >
> > Can you insert a comment here?
> >
> > # GCC 10 renamed --param=allow-store-data-races=0 to --allow-store-data-races
> >
> >
> > It will remind us of dropping this conditional
> > in the (long long distant) future.
> >
> >
> >
> >
> > > +KBUILD_CFLAGS  += $(call cc-option,--allow-store-data-races,--param=allow-store-data-races=0)
> > >  KBUILD_CFLAGS  += $(call cc-option,-fno-allow-store-data-races)
> > > +endif
>
> This report is confusing:
> https://lore.kernel.org/linux-mm/202108160729.Lx0IJzq3-lkp@xxxxxxxxx/
> (csky gcc-11)
>
> >> csky-linux-gcc: error: unrecognized command-line option '--param=allow-store-data-races=0'; did you mean '--allow-store-data-races'?
>
> I wonder if cc-option detection for these is broken?

I do not say it is broken...


cc-option is defined like this:

cc-option = $(call __cc-option, $(CC),\
        $(KBUILD_CPPFLAGS) $(KBUILD_CFLAGS),$(1),$(2))


It is checking
$(KBUILD_CPPFLAGS) + $(KBUILD_CFLAGS)
+ --allow-store-data-races


A few lines above, I see

csky-linux-gcc: error: unrecognized argument in option '-mcpu=ck860'


It makes all the cc-option tests fail after this line:
KBUILD_CFLAGS += -mcpu=$(CPUTYPE) -Wa,-mcpu=$(MCPU_STR)


Then,

$(call cc-option,--allow-store-data-races,--param=allow-store-data-races=0)

falls back to --param=allow-store-data-races=0




>  Perhaps I should
> not touch these other than to wrap them in the CONFIG_CC_IS_GCC guard?

I do not think so.

If an unrecognized argument appears,
all the cc-option tests that follow are unreliable.



If you are not comfortable with  it,
you can split it.

KBUILD_CFLAGS  += $(call cc-option,--allow-store-data-races)
KBUILD_CFLAGS  += $(call cc-option,--param=allow-store-data-races=0)



Or, another way of implementation is

KBUILD_CFLAGS += $(call cc-ifversion, -lt, 1000,
--allow-store-data-races, --param=allow-store-data-races=0)





>
> (Either way, I need to send a v2 in response to Naresh's report as
> well. https://lore.kernel.org/lkml/CA+G9fYtPBp_-Ko_P7NuOX6vN9-66rjJuBt21h3arrLqEaQQn6w@xxxxxxxxxxxxxx/
> It seems that -mfentry wasn't implemented for s390-linux-gnu-gcc until
> gcc-9; so rather than remove top level support, perhaps a comment
> about gcc-9+ s390 having support will make grepping for it easier in
> the future).
> --
> Thanks,
> ~Nick Desaulniers



-- 
Best Regards
Masahiro Yamada