Re: [PATCH 1/4] kbuild: remove LLVM=1 test from HAS_LTO_CLANG

From: Sami Tolvanen <samitolvanen@google.com>
To: Nick Desaulniers <ndesaulniers@google.com>
Cc: Masahiro Yamada <masahiroy@kernel.org>,
	Linux Kbuild mailing list <linux-kbuild@vger.kernel.org>,
	clang-built-linux <clang-built-linux@googlegroups.com>,
	Nathan Chancellor <nathan@kernel.org>,
	LKML <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH 1/4] kbuild: remove LLVM=1 test from HAS_LTO_CLANG
Date: Mon, 8 Mar 2021 11:11:28 -0800	[thread overview]
Message-ID: <CABCJKudmzK=AhtMpZt6X3BDG7NHmf9C6ncEt9JJAv+uto14qXQ@mail.gmail.com> (raw)
In-Reply-To: <CAKwvOdkhZGv_q9vgDdYY44OrbzmMD_E+GL3SyOk-jQ0kdXtMzg@mail.gmail.com>

On Wed, Mar 3, 2021 at 12:47 PM Nick Desaulniers
<ndesaulniers@google.com> wrote:
>
> + Sami
>
> On Wed, Mar 3, 2021 at 10:34 AM Masahiro Yamada <masahiroy@kernel.org> wrote:
> >
> > This guarding is wrong. As Documentation/kbuild/llvm.rst notes, LLVM=1
> > switches the default of tools, but you can still override CC, LD, etc.
> > individually.
> >
> > BTW, LLVM is not 1/0 flag. If LLVM is not passed in, it is empty.
>
> Do we have the same problem with LLVM_IAS?  LGTM otherwise, but wanted
> to check that before signing off.
>
> (Also, the rest of the patches in this series seem more related to
> DWARFv5 cleanups; this patch seems orthogonal while those are a
> visible progression).
>
> >
> > Non-zero return code is all treated as failure anyway.
> >
> > So, $(success,test $(LLVM) -eq 1) and $(success,test "$(LLVM)" = 1)
> > works equivalently in the sense that both are expanded to 'n' if LLVM
> > is not given. The difference is that the former internally fails due
> > to syntax error.
> >
> >   $ test ${LLVM} -eq 1
> >   bash: test: -eq: unary operator expected
> >   $ echo $?
> >   2
> >
> >   $ test "${LLVM}" -eq 1
> >   bash: test: : integer expression expected
> >   $ echo $?
> >   2
> >
> >   $ test "${LLVM}" = 1
> >   echo $?
> >   1
> >
> >   $ test -n "${LLVM}"
> >   $ echo $?
> >   1
> >
> > Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
> > ---
> >
> >  arch/Kconfig | 1 -
> >  1 file changed, 1 deletion(-)
> >
> > diff --git a/arch/Kconfig b/arch/Kconfig
> > index 2bb30673d8e6..2af10ebe5ed0 100644
> > --- a/arch/Kconfig
> > +++ b/arch/Kconfig
> > @@ -632,7 +632,6 @@ config HAS_LTO_CLANG
> >         def_bool y
> >         # Clang >= 11: https://github.com/ClangBuiltLinux/linux/issues/510
> >         depends on CC_IS_CLANG && CLANG_VERSION >= 110000 && LD_IS_LLD
> > -       depends on $(success,test $(LLVM) -eq 1)
>
> IIRC, we needed some other LLVM utilities like llvm-nm and llvm-ar,
> which are checked below. So I guess we can still support CC=clang
> AR=llvm-ar NM=llvm-nm, and this check is redundant.

I'm fine with removing the check, but the idea here was to just make
it slightly harder for people to accidentally use a mismatched
toolchain, even though checking for LLVM=1 doesn't stop them from
doing so anyway. But yes, the only LLVM tools required in addition to
the compiler and the linker are llvm-ar and llvm-nm.

Sami