Re: The trouble with __weak and objtool got worse

[Fangrui Song <maskray@google.com>]

On 2022-04-15, Nick Desaulniers wrote:
>On Fri, Apr 15, 2022 at 1:11 PM Segher Boessenkool
><segher@kernel.crashing.org> wrote:
>>
>> On Fri, Apr 15, 2022 at 11:36:32AM -0700, Nick Desaulniers wrote:
>> > On Fri, Apr 15, 2022 at 11:27 AM Segher Boessenkool
>> > <segher@kernel.crashing.org> wrote:
>> > >
>> > > > Alternatively:
>> > > >
>> > > >   https://sourceware.org/pipermail/binutils/2020-December/114671.html
>> > > >
>> > > > seems to suggest: -Wa,--generate-unused-section-symbols=yes, ought to
>> > > > work, except I'm getting:
>> > >
>> > > That email is for a proposed patch.  Did anything further ever happen
>> > > with it?
>> >
>> > $ gcc hello.c -Wa,--generate-unused-section-symbols=yes
>> > as: unrecognized option '--generate-unused-section-symbols=yes'
>> > $ gcc hello.c -Wa,-generate-unused-section-symbols=yes
>> > $ gcc --version
>> > gcc (Debian 11.2.0-16) 11.2.0
>> >
>> > Uh, the email says -- prefix, but reality shows a single prefix? What
>> > happened there? Maybe that link is to an earlier version than what
>> > landed?
>>
>> Neither has landed.  You get the -g option, which can take options of
>> itself, parsed by the target or file format code.  "as -gobbledygook"
>> works fine as well :-)
>
>https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=d1bcae833b32f1408485ce69f844dcd7ded093a8
>Looks like it's been in binutils since 2.36 (IIUC the ChangeLog correctly).

(Came to this thread from https://reviews.llvm.org/D123874)

H.J.'s patch discards unused STT_SECTION symbols.

The ability to add back such symbols (--generate-unused-section-symbols={yes|no}) is not in binutils.
Neither 2.36 nor master branch has --generate-unused-section-symbols={yes|no}.

as -generate-unused-section-symbols=yes is just a GNU as quirk that it
interprets nearly all -g* options as -g, which generates .debug_arange
and .debug_info compile units. The feature is very different from
restoring STT_SECTION symbols.

I know little about the kernel requirement but I think it shouldn't be
too difficult to teach a relocatable object file consumer to not assume
existence of STT_SECTION. The converse (add the feature to GNU as and
LLVM integrated assembler) would be much more unideal.

See also
https://sourceware.org/pipermail/binutils/2022-March/119940.html

> This problem was fixed in the kernel (and backported by distros to their 
> kernels if binutils was updated); it's fairly simplistic changes.
> 
> I don't see the need for returning (optionally) to the old behaviour 
> again.  I mean, at the time, when 2.36 came out, sure, the patch might 
> have made sense, but now?