[PATCH] ld/ELF: explicitly place .note.gnu.property

[Matthieu Longo]

On 2025-08-18 08:08, Jan Beulich wrote:
> On 15.08.2025 17:03, Matthieu Longo wrote:
>> On 2025-08-15 13:23, Jan Beulich wrote:
>>> x86 gas may produce .note.gnu.property by default. Hence the default
>>> linker script would better also deal with that section, rather than
>>> causing diagnostics when using --orphan-handling=. As the origin of the
>>> (seemingly mis-spelled) .note.gnu-property there is unclear, leave that
>>> in place.
>>> ---
>>> Furthermore, the comment ahead of this block of directives has gone
>>> stale with, afaict, the placement of .note.gnu.build-id near the
>>> program headers: Orphan note sections are placed adjacent to that one if
>>> --build-id= is in use. This pretty certainly wasn't intended, so may
>>> also want/need fixing.
>>>
>>> The testsuite adjustment is likely wrong to make, but I can't seem to
>>> be able to figure what's going on for powerpc64-linux: .eh_frame,
>>> despite being empty as before, suddenly has SEC_KEEP set (by
>>> gc_section_callback() as invoked from lang_gc_sections_1(), in
>>> accordance with the linker script using KEEP() on the section), and
>>> hence maybe_strip_output() wouldn't remove it anymore. It thus
>>> unexpectedly appears as constituent of the GNU_PROPERTY segment.
>>>
>>> Whether the two Arm64 linker tests that use .note.gnu-property are also
>>> screwed I can't easily tell. The linker script fragment (back then in
>>> miscsections.sc) appeared "out of the blue" (7ecf0250f7f0 ["Improve the
>>> placement of orphan note sections"]), so whether some target uses that
>>> section name isn't quite easy to tell. Similarly the addition to the two
>>> testsuite linker scripts appeared with no explanation at all
>>> (1905f68578fa ["AArch64 BTI/PAC PLT tests: remove RWX permissions on
>>> segments"]).
>>>
>>> --- a/ld/scripttempl/elf.sc
>>> +++ b/ld/scripttempl/elf.sc
>>> @@ -735,6 +735,7 @@ cat <<EOF
>>>      .note.build-id :      { *(.note.build-id) } ${RELOCATING+${REGION}}
>>>      .note.GNU-stack :     { *(.note.GNU-stack) } ${RELOCATING+${REGION}}
>>>      .note.gnu-property :  { *(.note.gnu-property) } ${RELOCATING+${REGION}}
>>> +  .note.gnu.property :  { *(.note.gnu.property) } ${RELOCATING+${REGION}}
>>>      .note.ABI-tag :       { *(.note.ABI-tag) } ${RELOCATING+${REGION}}
>>>      .note.package :       { *(.note.package) } ${RELOCATING+${REGION}}
>>>      .note.dlopen :        { *(.note.dlopen) } ${RELOCATING+${REGION}}
>>> --- a/ld/testsuite/ld-elf/elf.exp
>>> +++ b/ld/testsuite/ld-elf/elf.exp
>>> @@ -86,6 +86,9 @@ run_ld_link_tests [list \
>>>    
>>>    if [is_elf64 tmpdir/symbol3w.a] {
>>>        set ASFLAGS "$ASFLAGS --defsym ALIGN=3"
>>> +    if { [istarget powerpc*-*-linux*] } {
>>> +	set LDFLAGS "$LDFLAGS --no-ld-generated-unwind-info"
>>> +    }
>>>        set pr23900_1_exp "pr23900-1-64.rd"
>>>        set pr25490_2_exp "pr25490-2-64.rd"
>>>        set pr25490_3_exp "pr25490-3-64.rd"
>>
>> The two occurrences of ".note.gnu-property" in the linker scripts used
>> by the tests is a misspelling.
>> Please feel free to fix it as a part of this patch.
> 
> Yet then I wonder: Does this section need mentioning at all in those two
> linker scripts?
> 
> Jan

The section ".note.gnu.property" is generated by the tests. It seems to 
work without the mention in the linker script, but I would prefer to 
keep it explicitly mentioned there.

Matthieu