Received-SPF: pass (google.com: domain of linux-kernel+bounces-176884-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
Precedence: bulk
MIME-Version: 1.0
References: <Zj9UDiAHqX3noTsy@archie.me> <CAK7LNARrG0u2pSzAkdbzrcf_i7ByhEnDwHH=4LtaE_H7OgL=0A@mail.gmail.com>
In-Reply-To: <CAK7LNARrG0u2pSzAkdbzrcf_i7ByhEnDwHH=4LtaE_H7OgL=0A@mail.gmail.com>
From: Masahiro Yamada <masahiroy@kernel.org>
Date: Sun, 12 May 2024 22:14:40 +0900
Message-ID: <CAK7LNASVN7he-iuSTY8Z2cvxRA=97iqZHbW3mNL_38xzdbcvyw@mail.gmail.com>
Subject: Re: Incorrect thinlto cache directory path in Makefile
To: Bagas Sanjaya <bagasdotme@gmail.com>
Cc: Linux Kernel Mailing List <linux-kernel@vger.kernel.org>, 
	Linux Kernel Build System <linux-kbuild@vger.kernel.org>, Sami Tolvanen <samitolvanen@google.com>, 
	Kees Cook <keescook@chromium.org>, Xu Zhen <xuzhen@fastmail.com>, 
	Nathan Chancellor <nathan@kernel.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Sun, May 12, 2024 at 10:04=E2=80=AFPM Masahiro Yamada <masahiroy@kernel.=
org> wrote:
>
> +CC: Nathan
>
>
> On Sat, May 11, 2024 at 8:18=E2=80=AFPM Bagas Sanjaya <bagasdotme@gmail.c=
om> wrote:
> >
> > Hi,
> >
> > Xu Zhen <xuzhen@fastmail.com> reported on Bugzilla (https://bugzilla.ke=
rnel.org/show_bug.cgi?id=3D218825) thinlto build directory problem (especia=
lly on
> > DKMS modules). He wrote:
> >
> > > In Makefile, the ld flag `--thinlto-cache-dir' is set to
> > > `$(extmod_prefix).thinlto-cache'. But at that time, the variable extm=
od_prefix
> > > had not been assigned yet. Therefore, the thinlto cache dir is always=
 created
> > > in the current directory.
> > >
> > > Even worse, the cache dir cannot be deleted when executing `make clea=
n'. This
> > > is because its path was written as `$(KBUILD_EXTMOD)/.thinlto-cache' =
in the
> > > clean: target.
> > >
> > > Some users have been troubled by it: https://github.com/dell/dkms/iss=
ues/292
> > >
> > > This bug was introduced by commit dc5723b02e523b2c4a68667f7e28c65018f=
7202f
>
>
> I agree this is a bug.
>
>
>
>
> line 945:
>   KBUILD_LDFLAGS  +=3D --thinlto-cache-dir=3D$(extmod_prefix).thinlto-cac=
he
>
>
> line 1034:
>   KBUILD_LDFLAGS  +=3D $(call ld-option,--no-warn-rwx-segments)


This one is probably OK because it is guarded by CONFIG_LD_IS_BFD,
while LTO requires LD_IS_LLD.


I see another ld-option at 1043.

  LDFLAGS_vmlinux +=3D $(call ld-option,--pack-dyn-relocs=3Drelr,-z
pack-relative-relocs)


Maybe, this is related to the issue
(although I did not attempt to reproduce it).


--=20
Best Regards
Masahiro Yamada