[PATCH] s390: Add linker tests for references in discarded sections
Jens Remus
jremus@linux.ibm.com
Thu Nov 12 12:09:19 GMT 2026
Hello Andreas,
thank you for the review and feedback!
Meanwhile I realized that I should have done better research: This
started out as porting the aarch64 relr-discard-* tests for my WIP
s390 support for DT_RELR, where the PIE test ran into the assertion.
I then ported the related RISC-V discard-* tests which confirmed the
PIE tests would also fail. Remembering my WIP work to port the x86-64
split of check_relocs into check_relocs and scan_relocs I verified that
both fails and the superfluous NONE and GLOB_DAT relocations would go
away. So I intended to use this patch to document the underlying issue
is unrelated to DT_RELR and would get resolved by the check_relocs
split.
I now looked into the RISC-V commit that introduced the tests (which I
should have done first) that fixes the issue that triggers the
assertion (not the superfluous NONE and GLOB_DAT relocations) and also
found the likewise x86-64 commit.
The tests from this patch are now part of my new patch that ports the
x86-64 fix, that I just sent for review:
[PATCH] s390: Do not force symbol dynamic unless it is undefined weak
https://inbox.sourceware.org/binutils/20261112115654.1527835-2-jremus@linux.ibm.com/
Regards,
Jens
On 11/5/2026 12:48 PM, Andreas Krebbel wrote:
> Hi Jens,
>
> this ok, thanks!
>
> Just two minor nits. Feel free to ignore:
>
> - discard-pie.d doesn't seem to include the expected output. So it would not turn into an XPASS automatically in case the problem goes away, right?
> - Should we, in discard-shared.d, document that the outcome is not what we want but rather what it is today?
>
> Andreas
>
>
> On 10/2/26 4:52 PM, Jens Remus wrote:
>
>> Add test cases verifying the behavior when linking with references to
>> symbols in discarded sections.
>>
>> Linking as PIE with a discarded section that contains a GOT reference
>> triggers a BFD assertion ((h->got.offset & 1) != 0) because the GOT
>> entry is allocated but never initialized. Mark the PIE test as XFAIL.
>> The PDE test passes because no dynamic relocations are needed. The
>> shared library test passes but documents the spurious R_390_NONE
>> relocations emitted for GOT and data references in discarded sections.
>>
>> Inspired by aarch64 relr-discard-* and riscv discard-* tests.
>>
>> ld/testsuite/
>> * ld-s390/s390.exp: Run new discard-* tests.
>> * ld-s390/discard.ld: New linker script.
>> * ld-s390/discard.s: New test source.
>> * ld-s390/discard-pde.d: New test for discarded sections in
>> a position-dependent executable (PDE).
>> * ld-s390/discard-pie.d: New test for discarded sections in
>> a position-independent executable (PIE). (XFAIL)
>> * ld-s390/discard-shared.d: New test for discarded sections in
>> a shared library.
>>
>> Signed-off-by: Jens Remus <jremus@linux.ibm.com>
>> ---
>>
>> Notes (jremus):
>> The plan is to address the XFAIL and superfluous R_390_NONE
>> relocations with a future patch, implementing x86-64's solution
>> using check_relocs and scan_relocs from x86 commit 3747999c6bd4
>> ("ld: Extract _bfd_elf_link_iterate_on_relocs").
>>
>> ld/testsuite/ld-s390/discard-pde.d | 6 +++
>> ld/testsuite/ld-s390/discard-pie.d | 13 ++++++
>> ld/testsuite/ld-s390/discard-shared.d | 13 ++++++
>> ld/testsuite/ld-s390/discard.ld | 14 ++++++
>> ld/testsuite/ld-s390/discard.s | 61 +++++++++++++++++++++++++++
>> ld/testsuite/ld-s390/s390.exp | 6 +++
>> 6 files changed, 113 insertions(+)
>> create mode 100644 ld/testsuite/ld-s390/discard-pde.d
>> create mode 100644 ld/testsuite/ld-s390/discard-pie.d
>> create mode 100644 ld/testsuite/ld-s390/discard-shared.d
>> create mode 100644 ld/testsuite/ld-s390/discard.ld
>> create mode 100644 ld/testsuite/ld-s390/discard.s
--
Jens Remus
Linux on Z Development (D3303)
jremus@de.ibm.com / jremus@linux.ibm.com
IBM Deutschland Research & Development GmbH; Vorsitzender des Aufsichtsrats: Wolfgang Wendt; Geschäftsführung: David Faller; Sitz der Gesellschaft: Ehningen; Registergericht: Amtsgericht Stuttgart, HRB 243294
IBM Data Privacy Statement: https://www.ibm.com/privacy/
More information about the Binutils
mailing list