Re: [PATCH 4/4] Update tests for protected data and function symbols

["H.J. Lu" <hjl.tools@gmail.com>]

On Mon, Jun 21, 2021 at 5:57 AM Florian Weimer <fweimer@redhat.com> wrote:
>
> * H. J. Lu:
>
> > On Mon, Jun 21, 2021 at 12:19 AM Florian Weimer <fweimer@redhat.com> wrote:
> >>
> >> * H. J. Lu via Libc-alpha:
> >>
> >> > Protected data and function symbols don't work well without
> >> > -fsingle-global-definition:
> >> >
> >> > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=37611
> >> > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=44166
> >> >
> >> > 1. Compile tst-protected1[ab].c and tst-protected1mod[ab].c with
> >> > -fsingle-global-definition if possible so that GOT entries are used
> >> > for undefined data accesses.
> >> > 2. Add tests for protected function pointers.
> >> > 3. Build tst-prelink.c without single global definition to keepp COPY
> >> > relocation.
> >>
> >> I think these tests need to check that the statically linked bits from
> >> GCC have been built with -fsingle-global-definition.  I don't think
> >> that's guaranteed even if GCC supports -fsingle-global-definition.
> >>
> >> I think this shows the limitation of the single bit of markup: the
> >> statically linked GCC bits are all hidden, so setting
> >> GNU_PROPERTY_1_NEEDED_SINGLE_GLOBAL_DEFINITION so is not correct as a
> >> marker to *require* a single global definition, but neither is not
> >> setting it because libgcc.a etc. should be *compatible* with
> >> GNU_PROPERTY_1_NEEDED_SINGLE_GLOBAL_DEFINITION.
> >
> > The output bit is ORed from all inputs.   When one input object file doesn't
> > work with COPY relocation, the final output object is marked incompatible
> > with COPY relocation even if some inputs do work with COPY relocation.
>
> But for the main executable, it has to work the other way round.
>

We need a new property which is ANDed for executables.

-- 
H.J.