Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <478f608e3b8de1218798c792b34dca75fa91f6a9.camel@gmail.com>
Subject: Re: [PATCH bpf] bpf/btf: Accept function names that contain dots
From:   Eduard Zingerman <eddyz87@gmail.com>
To:     Florent Revest <revest@chromium.org>
Cc:     bpf@vger.kernel.org, linux-kernel@vger.kernel.org,
        llvm@lists.linux.dev, martin.lau@linux.dev, ast@kernel.org,
        daniel@iogearbox.net, andrii@kernel.org, song@kernel.org,
        yhs@fb.com, john.fastabend@gmail.com, kpsingh@kernel.org,
        sdf@google.com, haoluo@google.com, jolsa@kernel.org,
        nathan@kernel.org, ndesaulniers@google.com, trix@redhat.com,
        stable@vger.kernel.org
Date:   Mon, 19 Jun 2023 18:24:49 +0300
In-Reply-To: <CABRcYmKY_4_udQtsu7E9CVPruPphnejcgCvGnfHzzu-yc4Kshg@mail.gmail.com>
References: <20230615145607.3469985-1-revest@chromium.org>
         <CABRcYm+C+tPwXAGnaDRR_U2hzyt+09fjkKBp3tPx6iKT4wBE2Q@mail.gmail.com>
         <fbd79f5f2b250ec913c78d91b94ca96fb96f67ee.camel@gmail.com>
         <CABRcYmLaummOg=Nf0qXVN2eci=25OqXLD0zpCUz4CgmTjvo9LA@mail.gmail.com>
         <CABRcYmKY_4_udQtsu7E9CVPruPphnejcgCvGnfHzzu-yc4Kshg@mail.gmail.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.44.4-0ubuntu1 
MIME-Version: 1.0
Precedence: bulk

On Mon, 2023-06-19 at 15:55 +0200, Florent Revest wrote:
> On Mon, Jun 19, 2023 at 1:20=E2=80=AFPM Florent Revest <revest@chromium.o=
rg> wrote:
> >=20
> > On Thu, Jun 15, 2023 at 7:05=E2=80=AFPM Eduard Zingerman <eddyz87@gmail=
.com> wrote:
> > >=20
> > > On Thu, 2023-06-15 at 17:44 +0200, Florent Revest wrote:
> > > > An easy reproducer is:
> > > >=20
> > > > $ touch pwet.c
> > > >=20
> > > > $ clang -g -fsanitize=3Dkernel-address -c -o pwet.o pwet.c
> > > > $ llvm-dwarfdump pwet.o | grep module_ctor
> > > >=20
> > > > $ clang -fno-integrated-as -g -fsanitize=3Dkernel-address -c -o pwe=
t.o pwet.c
> > > > $ llvm-dwarfdump pwet.o | grep module_ctor
> > > >                 DW_AT_name      ("asan.module_ctor")
> > >=20
> > > Interestingly, I am unable to reproduce it using either
> > > clang version 14.0.0-1ubuntu1 or clang main (bd66f4b1da30).
> >=20
> > Somehow, I didn't think of trying other clang versions! Thanks, that's
> > a good point Eduard. :)
> >=20
> > I also can't reproduce it on a 14x build.
> >=20
> > However, I seem to be able to reproduce it on main:
> >=20
> >   git clone https://github.com/llvm/llvm-project.git
> >   mkdir llvm-project/build
> >   cd llvm-project/build
> >   git checkout bd66f4b1da30
> >   cmake -DLLVM_ENABLE_PROJECTS=3Dclang -DCMAKE_BUILD_TYPE=3DRelease -G
> > "Unix Makefiles" ../llvm
> >   make -j $(nproc)
> >=20
> >   bin/clang -fno-integrated-as -g -fsanitize=3Dkernel-address -c -o
> > ~/pwet.o ~/pwet.c
> >   bin/llvm-dwarfdump ~/pwet.o | grep module_ctor
> >   # Shows module_ctor
> >=20
> > I started a bisection, hopefully that will point to something interesti=
ng
>=20
> The bisection pointed to a LLVM patch from Nick in October 2022:
> e3bb359aacdd ("[clang][Toolchains][Gnu] pass -g through to assembler")
>=20
> Based on the context I have, that commit sounds fair enough. I don't
> think LLVM does anything wrong here, it seems like BPF should be the
> one dealing with dots in function debug info.

That explains why I could not reproduce the issue: I tried with gas 2.38.
Using gas 2.40 I see the same behavior as you.

If one tries to generate assembly file with '-fsanitize':

  $ clang -fno-integrated-as -g -fsanitize=3Dkernel-address -S -o pwet.s pw=
et.c
  $ cat pwet.s
  	.text
  	.file	"pwet.c"
  	.p2align	4, 0x90                         # -- Begin function asan.module=
_ctor
  	.type	asan.module_ctor,@function
  asan.module_ctor:                       # @asan.module_ctor
  .Lfunc_begin0:
      ...

And then compile it using Gnu assembler:

  $ as --64 -o pwet.o pwet.s -g -gdwarf-5

The behavior differs between 2.38 and 2.40, the older version does not
produce debug entry for 'asan.module_ctor', while newer does.