Re: [PATCH] arm64: Kconfig: Enable MCT timer for ARCH_EXYNOS

[Sam Protsenko]

On Fri, 29 Oct 2021 at 14:38, Marek Szyprowski <m.szyprowski@xxxxxxxxxxx> wrote:
>
> On 28.10.2021 16:35, Krzysztof Kozlowski wrote:
> > On 28/10/2021 16:22, Sam Protsenko wrote:
> >> On Tue, 26 Oct 2021 at 17:03, Krzysztof Kozlowski
> >> <krzysztof.kozlowski@xxxxxxxxxxxxx> wrote:
> >>> On 26/10/2021 13:59, Sam Protsenko wrote:
> >>>> Some ARM64 Exynos SoCs have MCT timer block, e.g. Exynos850 and
> >>>> Exynos5433. CLKSRC_EXYNOS_MCT option is not visible unless COMPILE_TEST
> >>>> is enabled. Select CLKSRC_EXYNOS_MCT option for ARM64 ARCH_EXYNOS like
> >>>> it's done in arch/arm/mach-exynos/Kconfig, to enable MCT timer support
> >>>> for ARM64 Exynos SoCs.
> >>>>
> >>>> Signed-off-by: Sam Protsenko <semen.protsenko@xxxxxxxxxx>
> >>>> ---
> >>>>   arch/arm64/Kconfig.platforms | 1 +
> >>>>   1 file changed, 1 insertion(+)
> >>>>
> >>> +CC Marek, Marc, Mark and Chanwoo,
> >>> Looks like duplicated:
> >>> https://lore.kernel.org/lkml/20181018095708.1527-7-m.szyprowski@xxxxxxxxxxx/
> >>>
> >>> The topic stalled and I think this particular patch did not make sense
> >>> on its own, without rest of changes from Marek. I am not sure, though...
> >>>
> >> Krzysztof, Marek,
> >>
> >> That series looks nice, I'm quite interested in that being applied. Do
> >> you think I can do something to help with that (e.g. rebasing,
> >> re-sending on behalf of Marek, testing on Exynos850, etc)?
> > I think there were no objections against v4 of this patchset, but
> > somehow it wasn't applied.
> >
> > Marek,
> > Does it make sense to try respinning your v4?
>
> I think I've abandoned it, because I got a final NACK on the arch timer
> change (support for 'not-fw-configured' timers). Without that the above
> mentioned changes doesn't make much sense.
>
> I know that the Exynos5433 firmware breaks the defined protocol, but on
> the other hand I can do nothing more than carrying internally those few
> patches out of tree to keep it fully working with 'mainline'.
>
> I've observed the same problem with newer Exynos SoCs, but those so far
> didn't get mainline support (yet), although there have been some
> attempts from the community.
>

There is no such problem on Exynos850 SoC (which I'm trying to
upstream right now), the architectured timer seems to be working fine
there, as is. Or maybe everything is configured properly in
bootloader, not sure. Also I managed to use MCT as a primary
clocksource (for sched_clock, etc) by disabling arch timer in
arch/arm64/kernel/time.c (just for the sake of test), and it works
fine.  So I still would like to have MCT enabled in kernel: this way
we can at least test the driver, as the clocksource can be changed to
MCT via sysfs (or by disabling arch timer in time.c). And for that we
need to at least enable CLKSRC_EXYNOS_MCT in ARCH_EXYNOS.

I guess all your patches (except one you mentioned) from that series
should be fine to apply. Do you mind if I re-send your patch series
(minus one patch) on your behalf? That would spare you some boring
work, and at least minimize your local delta you're carrying.

Thanks!

> Best regards
>
> --
> Marek Szyprowski, PhD
> Samsung R&D Institute Poland
>