Re: [Patch v2] x86/build: require only gcc use -maccumulate-outgoing-args

[Josh Poimboeuf <jpoimboe@redhat.com>]

On Thu, May 04, 2017 at 11:38:57PM -0700, hpa@zytor.com wrote:
> On May 4, 2017 11:23:33 PM PDT, Ingo Molnar <mingo@kernel.org> wrote:
> >
> >* Nick Desaulniers <nick.desaulniers@gmail.com> wrote:
> >
> >> Other compilers, like clang, treat unknown compiler flags as errors.
> >> 
> >> Signed-off-by: Nick Desaulniers <nick.desaulniers@gmail.com>
> >> ---
> >>  arch/x86/Makefile | 2 +-
> >>  1 file changed, 1 insertion(+), 1 deletion(-)
> >> 
> >> diff --git a/arch/x86/Makefile b/arch/x86/Makefile
> >> index 4430dd489620..12757a252e6b 100644
> >> --- a/arch/x86/Makefile
> >> +++ b/arch/x86/Makefile
> >> @@ -179,7 +179,7 @@ ifdef CONFIG_JUMP_LABEL
> >>  endif
> >>  
> >>  ifeq ($(ACCUMULATE_OUTGOING_ARGS), 1)
> >> -	KBUILD_CFLAGS += -maccumulate-outgoing-args
> >> +	KBUILD_CFLAGS += $(if $(filter
> >gcc,$(cc-name)),-maccumulate-outgoing-args)
> >>  endif
> >>  
> >>  # Stackpointer is addressed different for 32 bit and 64 bit x86
> >
> >The justification Josh gave for this pattern should be put into a
> >comment and into 
> >the changelog as well.
> >
> >Thanks,
> >
> >	Ingo
> 
> However, I don't think Josh's explanation is correct.  I am pretty
> sure it is a performance issue, not a correctness issue

Why wouldn't it be a correctness issue?  The option is needed in a few
cases (involving older versions of gcc and certain configs) to avoid
some bugs (see the Makefile for more details).

> and besides, a version of gcc that old won't be able to compile the
> kernel for other reasons, as evidenced by the fact that noone has
> complained about this option being mandatory.

Yeah.  Looking at the gcc source, the option has actually been around
since 2000.  So, never mind!

I'd be ok with v1, plus a comment saying that clang doesn't support the
option.

-- 
Josh