Re: [PATCH 15/52] m68k: emu: Mark version[] __maybe_unused

[Arnd Bergmann]

On Sat, Sep 9, 2023, at 14:12, Geert Uytterhoeven wrote:
> Hi Arnd,
>
> On Sat, Sep 9, 2023 at 10:28 AM Arnd Bergmann <arnd@xxxxxxxx> wrote:
>> On Thu, Sep 7, 2023, at 15:41, Geert Uytterhoeven wrote:
>> > When building with W=1:
>> >
>> >     m68k/arch/m68k/emu/nfeth.c:42:19: warning: ‘version’ defined but
>> > not used [-Wunused-const-variable=]
>> >        42 | static const char version[] =
>> >         |                   ^~~~~~~
>> >
>> > Fix this while obeying the wishes of the original copyright holders by
>> > marking version[] with __maybe_unused.
>> >
>> > Signed-off-by: Geert Uytterhoeven <geert@xxxxxxxxxxxxxx>
>>
>> That will just drop it from the object file, maybe it should
>> instead be marked "__used" ?
>
> Given it's been like this since the beginning, I guess it never showed
> up in the object file anyway...

Older compilers used to keep unused variables, and all versions
still do when building with -O0 instead of the usual -O2.

I can see on godbolt.org that gcc-1.27 did not have that optimization
yet, while gcc-4.1 had it, but I haven't found any versions between
those. I vaguely remember us throwing out the version strings from
most kernel files a long time ago, possibly that was triggered by
gcc no longer putting them into the object any more, but I can't
find a reference for that.

What I can see is that a lot of old network drivers have version
strings like this one, but they are actually printed from the
probe function, and lost their __devinitdata annotation in the past.

      Arnd