[PATCH] vdso: Improve cmd_vdso_check to check all dynamic relocations

[Fangrui Song]

On Sun, Dec 4, 2022 at 8:58 AM Christophe Leroy
<christophe.leroy at csgroup.eu> wrote:
>
>
>
> Le 04/12/2022 à 05:50, Fangrui Song a écrit :
> > On Mon, Nov 14, 2022 at 4:46 PM Fangrui Song <maskray at google.com> wrote:
> >>
> >> On 2022-11-14, Christophe Leroy wrote:
> >>>
> >>>
> >>> Le 28/09/2022 à 07:25, Fangrui Song a écrit :
> >>>> On Sat, Sep 10, 2022 at 12:53 AM Fangrui Song <maskray at google.com> wrote:
> >>>>>
> >>>>> On 2022-08-30, Fangrui Song wrote:
> >>>>>> The actual intention is that no dynamic relocation exists. However, some
> >>>>>> GNU ld ports produce unneeded R_*_NONE. (If a port is not care enough to
> >>>>>> determine the exact .rel[a].dyn size, the trailing zeros become R_*_NONE
> >>>>>> relocations. E.g. powerpc64le ld as of 2.38 has the issue with
> >>>>>> defconfig.) R_*_NONE are generally no-op in the dynamic loaders. So just
> >>>>>> ignore them.
> >>>>>>
> >>>>>> With the change, we can remove ARCH_REL_TYPE_ABS. ARCH_REL_TYPE_ABS is a
> >>>>>> bit misnomer as ports may check RELAVETIVE/GLOB_DAT/JUMP_SLOT which are
> >>>>>> not called "absolute relocations". (The patch is motivated by the arm64
> >>>>>> port missing R_AARCH64_RELATIVE.)
> >>>>>>
> >>>>>> While here, replace "egrep" with "grep" as "egrep" is deprecated in GNU
> >>>>>> grep 3.7.
> >>>>>>
> >>>>>> Signed-off-by: Fangrui Song <maskray at google.com>
> >>>>>> ---
> >>>>>> [...]
> >>>>>>
> >>>>>
> >>>>> Ping.
> >>>>
> >>>> Ping^2 :)
> >>>
> >>> Can you explain which ARCH_REL_TYPE_ABS can be removed with this change ?
> >>> How is the verification done if ARCH_REL_TYPE_ABS is removed ?
> >>
> >> All ARCH_REL_TYPE_ABS relocation types can be removed. As explained, the
> >> real intention is to check no dynamic relocation, and this is done by
> >> the new
>
> Well, there was a typo in my question, I wanted to ask 'why', not 'which'.
>
> >>
> >>     cmd_vdso_check = if $(READELF) -rW $@ | grep -v _NONE | grep -q "R_\w*_"; \
> >>
> >> in this patch.
> >>
> >> grep -v _NONE is to work around some GNU ld ports (and likely older releases
> >> of some ports even if their latest versions are fixed) which produce
> >> unneeded R_*_NONE dynamic relocations.
> >
> > Ping :)
>
> Ok, that seems to work on powerpc.
>
> Reviewed-by: Christophe Leroy <christophe.leroy at csgroup.eu>

Thanks. Can a maintainer pick up this commit?