On 07/25/2025, Krzysztof Kozlowski wrote:
> On 25/07/2025 11:28, Daniel Lezcano wrote:
> > On 25/07/2025 11:03, Krzysztof Kozlowski wrote:
> >> Commit 5d86e479193b ("clocksource/drivers/exynos_mct: Add module
> >> support") introduced section mismatch failures.
> >> Commit 7e477e9c4eb4 ("clocksource/drivers/exynos_mct: Fix section
> >> mismatch from the module conversion") replaced these to other section
> >> mismatch failures:
> >>
> >> WARNING: modpost: vmlinux: section mismatch in reference: mct_init_dt+0x164 (section: .text) -> register_current_timer_delay (section: .init.text)
> >> WARNING: modpost: vmlinux: section mismatch in reference: mct_init_dt+0x20c (section: .text) -> register_current_timer_delay (section: .init.text)
> >> ERROR: modpost: Section mismatches detected.
> >>
> >> No progress on real fixing of these happened (intermediary fix was still
> >> not tested), so revert both commits till the work is prepared correctly.
> >
> > Please don't claim the fix was not tested. I reproduced the section
>
>
> section mismatch code MUST BE tested with enabled DEBUG_SECTION_MISMATCH
> and disabled SECTION_MISMATCH_WARN_ONLY. If you have warnings which you
> missed (although if you have warnings what did you fix?), means you did
> not prepare testing setup.
Thanks Krzysztof for reporting this! Sorry for the mess this has created. I was
unaware of testing with DEBUG_SECTION_MISMATCH and clearly missed those
warnings when I compile tested with ARCH=arm. I see the issue and will fix
the patches accordingly.
Regarding ARM32 runtime testing, how do I ensure my patch series is testing on
ARM32 devices? The series was discussed on the list for quite some time and I'm
wondering what step I missed to get this further tested? Do I just need to
explicitly request for ARM32 testing on the list?
To avoid this in the future, I took a look at the thread that introduced the
section mismatch warning [1] and wondering (before I propose this broadly) if
we can make section mismatches fatal by default for everything except
allmodconfig? Then we can selectively disable it where appropriate.
[1] https://lore.kernel.org/all/1443807963-36364-1-git-send-email-drinkcat@xxxxxxxxxxxx/
Thanks,
Will
>
> > mismatch, tested it and figured out it was indeed fixing the issue. I
> > just missed the error because it sounds very close to the first one
> > reported initially and I did the confusion.
> >
> > The driver is not supposed to be compiled as a module on ARM32.
> >
> > The option tristate "Exynos multi core timer driver" if ARM64 is
> > misleading. From this change, the defconfig on ARM can do
> > CONFIG_EXYNOS_MCT=m which should not be allowed.
> >
> > Before getting wild and revert everything, let's try to find a proper
> > fix for that.
>
> I am not wild here. The issue is there since 9 days.
>
> Best regards,
> Krzysztof
