On Tue, Sep 27, 2022 at 04:08:02PM -0300, Arnaldo Carvalho de Melo wrote:
> Em Tue, Sep 27, 2022 at 11:56:34AM -0700, Nathan Chancellor escreveu:
> > Hi Arnaldo,
> >
> > When building a kernel with LLVM and CONFIG_DEBUG_INFO_BTF after commit
> > 32ef9e5054ec ("Makefile.debug: re-enable debug info for .S files") in
> > the kernel, I see the following spew of warnings, which appear to come
> > from pahole:
> >
> > $ clang --version
> > clang version 15.0.0 (Fedora 15.0.0-3.fc38)
> > Target: x86_64-redhat-linux-gnu
> > Thread model: posix
> > InstalledDir: /usr/bin
> >
> > $ pahole --version
> > v1.24
> >
> > $ make -skj"$(nproc)" ARCH=x86_64 LLVM=1 defconfig
> >
> > $ scripts/config \
> > -d DEBUG_INFO_NONE \
> > -e BPF_SYSCALL \
> > -e DEBUG_INFO_BTF \
> > -e DEBUG_INFO_DWARF5
> >
> > $ make -skj"$(nproc)" ARCH=x86_64 LLVM=1 olddefconfig all
> > ...
> > die__process_unit: DW_TAG_label (0xa) @ <0x7b> not handled!
> > die__process_unit: tag not supported 0xa (label)!
> > die__process_unit: DW_TAG_label (0xa) @ <0x97> not handled!
> > die__process_unit: DW_TAG_label (0xa) @ <0xbd> not handled!
> > die__process_unit: DW_TAG_label (0xa) @ <0xed> not handled!
> > die__process_unit: DW_TAG_label (0xa) @ <0x109> not handled!
> > die__process_unit: DW_TAG_label (0xa) @ <0x12a> not handled!
> > die__process_unit: DW_TAG_label (0xa) @ <0x146> not handled!
> > die__process_unit: DW_TAG_label (0xa) @ <0x16f> not handled!
> > ...
> >
> > Is this a problem with LLVM or pahole? I do not see this when building
> > with GCC + GNU as but that could just be a red herring. I assume that
> > there could be something missing for processing debug info from
> > assembly, perhaps? If there is any further information I can provide or
> > anything I can test, I am more than happy to do so.
>
> I'll try to repro, but at first sight it looks like a label in a
If you need help reproducing this locally, feel free to reach out
through this thread or #clangbuiltlinux on Libera.
> DW_TAG_compile_unit/like level, I have to check what that means, but
> seems to be just a warning, did the end result made sense?
Right, this appears to just be a warning, the build still completes
successfully (I did not check much else). I only noticed this by
scrolling back in my build logs.
Cheers,
Nathan
