Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <CABCJKudYQV6Nt=Qq+zY=9JF1WmLVLnx+--3mJA9dkhLMhMvuFg@mail.gmail.com>
 <20210706090607.19421-1-lecopzer.chen@mediatek.com>
In-Reply-To: <20210706090607.19421-1-lecopzer.chen@mediatek.com>
From:   Sami Tolvanen <samitolvanen@google.com>
Date:   Wed, 7 Jul 2021 10:02:24 -0700
Message-ID: <CABCJKufWcp6Hx=8btz6pDNcKvQ21n4BSPZ7cp1Tzhxt0+pQOmw@mail.gmail.com>
Subject: Re: [PATCH v3 1/2] Kbuild: lto: add CONFIG_MAKE_VERSION
To:     Lecopzer Chen <lecopzer.chen@mediatek.com>
Cc:     clang-built-linux@googlegroups.com, keescook@chromium.org,
        linux-kbuild@vger.kernel.org, linux-kernel@vger.kernel.org,
        masahiroy@kernel.org, michal.lkml@markovi.net, nathan@kernel.org,
        ndesaulniers@google.com, yj.chiang@mediatek.com
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Jul 6, 2021 at 2:06 AM Lecopzer Chen <lecopzer.chen@mediatek.com> wrote:
>
> > On Sun, Jul 4, 2021 at 7:03 PM Masahiro Yamada <masahiroy@kernel.org> wrote:
> > >
> > > On Fri, Jul 2, 2021 at 12:29 PM Lecopzer Chen
> > > <lecopzer.chen@mediatek.com> wrote:
> > > >
> > > > To check the GNU make version. Used by the LTO Kconfig.
> > > >
> > > > LTO with MODVERSIONS will fail in generating correct CRC because
> > > > the makefile rule doesn't work for make with version 3.8X.[1]
> > > >
> > > > Thus we need to check make version during selecting on LTO Kconfig.
> > > > Add CONFIG_MAKE_VERSION which means MAKE_VERSION in canonical digits
> > > > for arithmetic comparisons.
> > > >
> > > > [1] https://lore.kernel.org/lkml/20210616080252.32046-1-lecopzer.chen@mediatek.com/
> > > > Signed-off-by: Lecopzer Chen <lecopzer.chen@mediatek.com>
> > > > ---
> > >
> > >
> > > NACK.
> > >
> > > "Let's add MAKE_VERSION >= 40200 restriction
> > > just because I cannot write correct code that
> > > works for older Make" is a horrible idea.
> > >
> > > Also, Kconfig is supposed to check the compiler
> > > (or toolchains) capability, not host tool versions.
> >
> > I feel like requiring a Make that's half a decade old for a feature
> > that also requires a toolchain released last October ago isn't
> > entirely unreasonable.
> >
> > That being said, if Masahiro prefers not to rely on the wildcard
> > function's behavior here, which is a reasonable request, we could
> > simply use the shell to test for the file's existence:
> >
> > diff --git a/scripts/Makefile.build b/scripts/Makefile.build
> > index 34d257653fb4..c6bd62f518ff 100644
> > --- a/scripts/Makefile.build
> > +++ b/scripts/Makefile.build
> > @@ -388,7 +388,7 @@ ifeq ($(CONFIG_LTO_CLANG) $(CONFIG_MODVERSIONS),y y)
> >        cmd_update_lto_symversions =                                     \
> >         rm -f $@.symversions                                            \
> >         $(foreach n, $(filter-out FORCE,$^),                            \
> > -               $(if $(wildcard $(n).symversions),                      \
> > +               $(if $(shell test -s $(n).symversions && echo y),       \
> >                         ; cat $(n).symversions >> $@.symversions))
> >  else
> >        cmd_update_lto_symversions = echo >/dev/null
> >
> > This is not quite as efficient as using wildcard, but should work with
> > older Make versions too. Thoughts?
> >
>
>
> I've tested this in both make-4.3 and 3.81, and the CRC is correct.
> But I'm not sure if anyone would have the "arg list too long" issue.
>
> Tested-by: Lecopzer Chen <lecopzer.chen@mediatek.com>

Thank you for testing. This should produce a command identical to the
wildcard version (with newer Make versions), so that shouldn't be an
issue. If nobody objects to this approach, would you mind putting this
into a proper patch and sending it as v4?

Sami