[PATCH] libbpf: fix readelf output parsing on powerpc with recent binutils

[Aurelien Jarno]

On 2019-12-11 09:33, Justin Forbes wrote:
> On Tue, Dec 10, 2019 at 4:26 PM Thadeu Lima de Souza Cascardo
> <cascardo at canonical.com> wrote:
> >
> > On Tue, Dec 10, 2019 at 12:58:33PM -0600, Justin Forbes wrote:
> > > On Mon, Dec 2, 2019 at 3:37 AM Daniel Borkmann <daniel at iogearbox.net> wrote:
> > > >
> > > > On Mon, Dec 02, 2019 at 04:53:26PM +1100, Michael Ellerman wrote:
> > > > > Aurelien Jarno <aurelien at aurel32.net> writes:
> > > > > > On powerpc with recent versions of binutils, readelf outputs an extra
> > > > > > field when dumping the symbols of an object file. For example:
> > > > > >
> > > > > >     35: 0000000000000838    96 FUNC    LOCAL  DEFAULT [<localentry>: 8]     1 btf_is_struct
> > > > > >
> > > > > > The extra "[<localentry>: 8]" prevents the GLOBAL_SYM_COUNT variable to
> > > > > > be computed correctly and causes the checkabi target to fail.
> > > > > >
> > > > > > Fix that by looking for the symbol name in the last field instead of the
> > > > > > 8th one. This way it should also cope with future extra fields.
> > > > > >
> > > > > > Signed-off-by: Aurelien Jarno <aurelien at aurel32.net>
> > > > > > ---
> > > > > >  tools/lib/bpf/Makefile | 4 ++--
> > > > > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > > >
> > > > > Thanks for fixing that, it's been on my very long list of test failures
> > > > > for a while.
> > > > >
> > > > > Tested-by: Michael Ellerman <mpe at ellerman.id.au>
> > > >
> > > > Looks good & also continues to work on x86. Applied, thanks!
> > >
> > > This actually seems to break horribly on PPC64le with binutils 2.33.1
> > > resulting in:
> > > Warning: Num of global symbols in sharedobjs/libbpf-in.o (32) does NOT
> > > match with num of versioned symbols in libbpf.so (184). Please make
> > > sure all LIBBPF_API symbols are versioned in libbpf.map.
> > >
> > > This is the only arch that fails, with x86/arm/aarch64/s390 all
> > > building fine.  Reverting this patch allows successful build across
> > > all arches.
> > >
> > > Justin
> >
> > Well, I ended up debugging this same issue and had the same fix as Jarno's when
> > I noticed his fix was already applied.
> >
> > I just installed a system with the latest binutils, 2.33.1, and it still breaks
> > without such fix. Can you tell what is the output of the following command on
> > your system?
> >
> > readelf -s --wide tools/lib/bpf/sharedobjs/libbpf-in.o | cut -d "@" -f1 | sed 's/_v[0-9]_[0-9]_[0-9].*//' | awk '/GLOBAL/ && /DEFAULT/ && !/UND/ {print $0}'
> >
> 
> readelf -s --wide tools/lib/bpf/sharedobjs/libbpf-in.o | cut -d "@"
> -f1 | sed 's/_v[0-9]_[0-9]_[0-9].*//' | awk '/GLOBAL/ && /DEFAULT/ &&
> !/UND/ {print $0}'
>    373: 00000000000141bc  1376 FUNC    GLOBAL DEFAULT    1 libbpf_num_possible_cpus [<localentry>: 8]
>    375: 000000000001869c   176 FUNC    GLOBAL DEFAULT    1 btf__free [<localentry>: 8]

It seems that in your case the localentry part is added after the symbol
name. That doesn't match what is done in upstream binutils:

https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=blob;f=binutils/readelf.c;hb=refs/heads/master#l11485

Which version of binutils are you using? It looks like your version has
been modified to workaround this exact issue.

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien at aurel32.net                 http://www.aurel32.net