Received-SPF: pass (google.com: domain of linux-kernel+bounces-214036-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
Date: Thu, 13 Jun 2024 14:40:19 -0700
From: Nathan Chancellor <nathan@kernel.org>
To: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: dwarves@vger.kernel.org,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	bpf@vger.kernel.org, Alan Maguire <alan.maguire@oracle.com>,
	Jiri Olsa <jolsa@kernel.org>, Jan Engelhardt <jengelh@inai.de>,
	Matthias Schwarzott <zzam@gentoo.org>,
	Viktor Malik <vmalik@redhat.com>,
	Eduard Zingerman <eddyz87@gmail.com>,
	Jan Alexander Steffens <heftig@archlinux.org>,
	Domenico Andreoli <cavok@debian.org>,
	Dominique Leuenberger <dimstar@opensuse.org>,
	Daniel Xu <dxu@dxuuu.xyz>, Yonghong Song <yonghong.song@linux.dev>,
	llvm@lists.linux.dev
Subject: Re: ANNOUNCE: pahole v1.27 (reproducible builds, BTF kfuncs)
Message-ID: <20240613214019.GA1423015@thelio-3990X>
References: <ZmjBHWw-Q5hKBiwA@x1>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <ZmjBHWw-Q5hKBiwA@x1>

Hi Arnaldo,

On Tue, Jun 11, 2024 at 06:26:53PM -0300, Arnaldo Carvalho de Melo wrote:
> 	The v1.27 release of pahole and its friends is out, supporting
> parallel reproducible builds and encoding kernel kfuncs in BTF, allowing
> tools such as bpftrace to enumerate the available kfuncs and obtain its
> function signatures and return types.

After commit f632e75 ("dwarf_loader: Add the cu to the cus list early,
remove on LSK_DELETE"), I (and others[1]) notice a crash when running
pahole on modules built with Clang when CONFIG_LTO_CLANG is enabled:

  $ curl -LSso .config https://gitlab.archlinux.org/archlinux/packaging/packages/linux/-/raw/main/config

  $ scripts/config -d LTO_NONE -e LTO_CLANG_THIN

  $ make -skj"$(nproc)" ARCH=x86_64 LLVM=1 olddefconfig vmlinux crypto/cast_common.ko
  make[3]: *** [scripts/Makefile.modfinal:59: crypto/cast_common.ko] Error 139

I've isolated this to the following commands using the files available
at [2] (these were built with LLVM 18 but I could reproduce it with LLVM
17 and LLVM 19, so it appears to impact a number of versions):

  $ tar -tf clang-lto-pahole-1.27-crash.tar.zst
  clang-lto-pahole-1.27-crash/
  clang-lto-pahole-1.27-crash/cast_common.mod.o
  clang-lto-pahole-1.27-crash/module.lds
  clang-lto-pahole-1.27-crash/cast_common.o
  clang-lto-pahole-1.27-crash/cast_common.ko.bak
  clang-lto-pahole-1.27-crash/vmlinux
  clang-lto-pahole-1.27-crash/cast_common.ko

  $ tar -axf clang-lto-pahole-1.27-crash.tar.zst

  $ cd clang-lto-pahole-1.27-crash

  $ LLVM_OBJCOPY="llvm-objcopy" pahole-1.26 -J -j --btf_features=encode_force,var,float,enum64,decl_tag,type_tag,optimized_func,consistent_func --lang_exclude=rust --btf_base vmlinux cast_common.ko

  $ cp cast_common.ko{.bak,}

  $ LLVM_OBJCOPY="llvm-objcopy" pahole-1.27 -J -j --btf_features=encode_force,var,float,enum64,decl_tag,type_tag,optimized_func,consistent_func --lang_exclude=rust --btf_base vmlinux cast_common.ko
  fish: Job 1, '...' terminated by signal SIGSEGV (Address boundary error)

If there is any more information I can provide or patches I can test, I
am more than happy to do so.

[1]: https://gitlab.archlinux.org/archlinux/packaging/packages/pahole/-/issues/1
[2]: https://1drv.ms/u/s!AsQNYeB-IEbqqC2F28JuLy__Q7Vd?e=KsraMU

Cheers,
Nathan