[PATCH] clocksource/drivers/exynos_mct: Use percpu interrupts only on ARM64

[PATCH] clocksource/drivers/exynos_mct: Use percpu interrupts only on ARM64

[Marek Szyprowski]

For some unknown reasons forcing percpu interrupts for local timers
breaks CPU hotplug for 'little' cores on legacy ARM 32bit Exynos based
machines (for example Exynos5422-based Odroid-XU3/XU4 boards). Use percpu
flag only when driver is compiled for newer ARM64 architecture.

Fixes: f3cec54ee3bf ("clocksource/drivers/exynos_mct: Set local timer interrupts as percpu")
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
---
 drivers/clocksource/exynos_mct.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/clocksource/exynos_mct.c b/drivers/clocksource/exynos_mct.c
index 62febeb4e1de..fece6bbc190e 100644
--- a/drivers/clocksource/exynos_mct.c
+++ b/drivers/clocksource/exynos_mct.c
@@ -603,7 +603,8 @@ static int __init exynos4_timer_interrupts(struct device_node *np,
 			if (request_irq(mct_irq,
 					exynos4_mct_tick_isr,
 					IRQF_TIMER | IRQF_NOBALANCING |
-					IRQF_PERCPU,
+					(IS_ENABLED(CONFIG_ARM64) ?
+					 IRQF_PERCPU : 0),
 					pcpu_mevt->name, pcpu_mevt)) {
 				pr_err("exynos-mct: cannot register IRQ (cpu%d)\n",
 									cpu);
-- 
2.34.1

Re: [PATCH] clocksource/drivers/exynos_mct: Use percpu interrupts only on ARM64

[Krzysztof Kozlowski]

On 27/08/2025 12:26, Marek Szyprowski wrote:
> For some unknown reasons forcing percpu interrupts for local timers
> breaks CPU hotplug for 'little' cores on legacy ARM 32bit Exynos based
> machines (for example Exynos5422-based Odroid-XU3/XU4 boards). Use percpu
> flag only when driver is compiled for newer ARM64 architecture.
> 
> Fixes: f3cec54ee3bf ("clocksource/drivers/exynos_mct: Set local timer interrupts as percpu")


I am pretty sure the patch above was not tested on arm32, thus this
workaround seems reasonable.

Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>



Best regards,
Krzysztof

Re: [PATCH] clocksource/drivers/exynos_mct: Use percpu interrupts only on ARM64

[Marek Szyprowski]

On 27.08.2025 13:58, Krzysztof Kozlowski wrote:
> On 27/08/2025 12:26, Marek Szyprowski wrote:
>> For some unknown reasons forcing percpu interrupts for local timers
>> breaks CPU hotplug for 'little' cores on legacy ARM 32bit Exynos based
>> machines (for example Exynos5422-based Odroid-XU3/XU4 boards). Use percpu
>> flag only when driver is compiled for newer ARM64 architecture.
>>
>> Fixes: f3cec54ee3bf ("clocksource/drivers/exynos_mct: Set local timer interrupts as percpu")
>
> I am pretty sure the patch above was not tested on arm32, thus this
> workaround seems reasonable.
>
> Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>

Daniel, any chance to get this merged?

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

Re: [PATCH] clocksource/drivers/exynos_mct: Use percpu interrupts only on ARM64

[Peter Griffin]

On Wed, 27 Aug 2025 at 11:26, Marek Szyprowski <m.szyprowski@samsung.com> wrote:
>
> For some unknown reasons forcing percpu interrupts for local timers
> breaks CPU hotplug for 'little' cores on legacy ARM 32bit Exynos based
> machines (for example Exynos5422-based Odroid-XU3/XU4 boards). Use percpu
> flag only when driver is compiled for newer ARM64 architecture.
>
> Fixes: f3cec54ee3bf ("clocksource/drivers/exynos_mct: Set local timer interrupts as percpu")
> Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
> ---

Reviewed-by: Peter Griffin <peter.griffin@linaro.org>

Re: [PATCH] clocksource/drivers/exynos_mct: Use percpu interrupts only on ARM64

[Daniel Lezcano]

Hi Marek,

On 19/09/2025 23:31, Marek Szyprowski wrote:
> On 27.08.2025 13:58, Krzysztof Kozlowski wrote:
>> On 27/08/2025 12:26, Marek Szyprowski wrote:
>>> For some unknown reasons forcing percpu interrupts for local timers
>>> breaks CPU hotplug for 'little' cores on legacy ARM 32bit Exynos based
>>> machines (for example Exynos5422-based Odroid-XU3/XU4 boards). Use percpu
>>> flag only when driver is compiled for newer ARM64 architecture.
>>>
>>> Fixes: f3cec54ee3bf ("clocksource/drivers/exynos_mct: Set local timer interrupts as percpu")
>>
>> I am pretty sure the patch above was not tested on arm32, thus this
>> workaround seems reasonable.
>>
>> Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
> 
> Daniel, any chance to get this merged?

The patch does not apply. However it is not your fault but a glitch in 
the previous pull request and the regression found in the Exynos MCT.

Apparently, Thomas had to drop the entire pull request while I thought 
he dropped only Will's changes.

A pull request with the previous changes for v6.17-rc1 and with the new 
changes for v6.18-rc1 is about to be emitted but without the Exynos MCT 
changes for the modularization. That means your change does not apply 
correctly anymore on the current changes.

-- 
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog

Re: [PATCH] clocksource/drivers/exynos_mct: Use percpu interrupts only on ARM64

[Marek Szyprowski]

Hi Daniel,

On 24.09.2025 10:41, Daniel Lezcano wrote:
>
> On 19/09/2025 23:31, Marek Szyprowski wrote:
>> On 27.08.2025 13:58, Krzysztof Kozlowski wrote:
>>> On 27/08/2025 12:26, Marek Szyprowski wrote:
>>>> For some unknown reasons forcing percpu interrupts for local timers
>>>> breaks CPU hotplug for 'little' cores on legacy ARM 32bit Exynos based
>>>> machines (for example Exynos5422-based Odroid-XU3/XU4 boards). Use 
>>>> percpu
>>>> flag only when driver is compiled for newer ARM64 architecture.
>>>>
>>>> Fixes: f3cec54ee3bf ("clocksource/drivers/exynos_mct: Set local 
>>>> timer interrupts as percpu")
>>>
>>> I am pretty sure the patch above was not tested on arm32, thus this
>>> workaround seems reasonable.
>>>
>>> Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
>>
>> Daniel, any chance to get this merged?
>
> The patch does not apply. However it is not your fault but a glitch in 
> the previous pull request and the regression found in the Exynos MCT.
>
> Apparently, Thomas had to drop the entire pull request while I thought 
> he dropped only Will's changes.
>
> A pull request with the previous changes for v6.17-rc1 and with the 
> new changes for v6.18-rc1 is about to be emitted but without the 
> Exynos MCT changes for the modularization. That means your change does 
> not apply correctly anymore on the current changes.

Thanks for the information. I've checked and 
https://git.kernel.org/pub/scm/linux/kernel/git/daniel.lezcano/linux.git/commit/?h=timers/drivers/next 
branch indeed doesn't contain the whole Exynos MCT modularization 
patchset, so the $subject patch (which was supposed to fix it) is not 
really needed.

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

Re: [PATCH] clocksource/drivers/exynos_mct: Use percpu interrupts only on ARM64

[William McVicker]

Hi Daniel,

On 09/24/2025, Daniel Lezcano wrote:
> 
> Hi Marek,
> 
> On 19/09/2025 23:31, Marek Szyprowski wrote:
> > On 27.08.2025 13:58, Krzysztof Kozlowski wrote:
> > > On 27/08/2025 12:26, Marek Szyprowski wrote:
> > > > For some unknown reasons forcing percpu interrupts for local timers
> > > > breaks CPU hotplug for 'little' cores on legacy ARM 32bit Exynos based
> > > > machines (for example Exynos5422-based Odroid-XU3/XU4 boards). Use percpu
> > > > flag only when driver is compiled for newer ARM64 architecture.
> > > > 
> > > > Fixes: f3cec54ee3bf ("clocksource/drivers/exynos_mct: Set local timer interrupts as percpu")
> > > 
> > > I am pretty sure the patch above was not tested on arm32, thus this
> > > workaround seems reasonable.
> > > 
> > > Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
> > 
> > Daniel, any chance to get this merged?
> 
> The patch does not apply. However it is not your fault but a glitch in the
> previous pull request and the regression found in the Exynos MCT.
> 
> Apparently, Thomas had to drop the entire pull request while I thought he
> dropped only Will's changes.
> 
> A pull request with the previous changes for v6.17-rc1 and with the new
> changes for v6.18-rc1 is about to be emitted but without the Exynos MCT
> changes for the modularization. That means your change does not apply
> correctly anymore on the current changes.

Is there somewhere I can check to see which changes will be re-sent? I don't
see them in the timers/drivers/next branch. Will it be patches 2-4 plus the
fixes for the section mismatch and this new IRQ fix? Once that's all sorted
out, I'll base the new modularization patches on top of that.

Thanks,
Will

Re: [PATCH] clocksource/drivers/exynos_mct: Use percpu interrupts only on ARM64

[Marek Szyprowski]

On 24.09.2025 19:17, William McVicker wrote:
> On 09/24/2025, Daniel Lezcano wrote:
>> On 19/09/2025 23:31, Marek Szyprowski wrote:
>>> On 27.08.2025 13:58, Krzysztof Kozlowski wrote:
>>>> On 27/08/2025 12:26, Marek Szyprowski wrote:
>>>>> For some unknown reasons forcing percpu interrupts for local timers
>>>>> breaks CPU hotplug for 'little' cores on legacy ARM 32bit Exynos based
>>>>> machines (for example Exynos5422-based Odroid-XU3/XU4 boards). Use percpu
>>>>> flag only when driver is compiled for newer ARM64 architecture.
>>>>>
>>>>> Fixes: f3cec54ee3bf ("clocksource/drivers/exynos_mct: Set local timer interrupts as percpu")
>>>> I am pretty sure the patch above was not tested on arm32, thus this
>>>> workaround seems reasonable.
>>>>
>>>> Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
>>> Daniel, any chance to get this merged?
>> The patch does not apply. However it is not your fault but a glitch in the
>> previous pull request and the regression found in the Exynos MCT.
>>
>> Apparently, Thomas had to drop the entire pull request while I thought he
>> dropped only Will's changes.
>>
>> A pull request with the previous changes for v6.17-rc1 and with the new
>> changes for v6.18-rc1 is about to be emitted but without the Exynos MCT
>> changes for the modularization. That means your change does not apply
>> correctly anymore on the current changes.
> Is there somewhere I can check to see which changes will be re-sent? I don't
> see them in the timers/drivers/next branch. Will it be patches 2-4 plus the
> fixes for the section mismatch and this new IRQ fix? Once that's all sorted
> out, I'll base the new modularization patches on top of that.

The problematic mct patches are still in linux-next, merged via tip tree 
by commit 9703240ef617 ("Merge branch into tip/master: 
'timers/clocksource'"). What are the plans to resolve this?

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

Re: [PATCH] clocksource/drivers/exynos_mct: Use percpu interrupts only on ARM64

[Daniel Lezcano]

On 25/09/2025 08:37, Marek Szyprowski wrote:
> On 24.09.2025 19:17, William McVicker wrote:
>> On 09/24/2025, Daniel Lezcano wrote:
>>> On 19/09/2025 23:31, Marek Szyprowski wrote:
>>>> On 27.08.2025 13:58, Krzysztof Kozlowski wrote:
>>>>> On 27/08/2025 12:26, Marek Szyprowski wrote:
>>>>>> For some unknown reasons forcing percpu interrupts for local timers
>>>>>> breaks CPU hotplug for 'little' cores on legacy ARM 32bit Exynos based
>>>>>> machines (for example Exynos5422-based Odroid-XU3/XU4 boards). Use percpu
>>>>>> flag only when driver is compiled for newer ARM64 architecture.
>>>>>>
>>>>>> Fixes: f3cec54ee3bf ("clocksource/drivers/exynos_mct: Set local timer interrupts as percpu")
>>>>> I am pretty sure the patch above was not tested on arm32, thus this
>>>>> workaround seems reasonable.
>>>>>
>>>>> Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
>>>> Daniel, any chance to get this merged?
>>> The patch does not apply. However it is not your fault but a glitch in the
>>> previous pull request and the regression found in the Exynos MCT.
>>>
>>> Apparently, Thomas had to drop the entire pull request while I thought he
>>> dropped only Will's changes.
>>>
>>> A pull request with the previous changes for v6.17-rc1 and with the new
>>> changes for v6.18-rc1 is about to be emitted but without the Exynos MCT
>>> changes for the modularization. That means your change does not apply
>>> correctly anymore on the current changes.
>> Is there somewhere I can check to see which changes will be re-sent? I don't
>> see them in the timers/drivers/next branch. Will it be patches 2-4 plus the
>> fixes for the section mismatch and this new IRQ fix? Once that's all sorted
>> out, I'll base the new modularization patches on top of that.
> 
> The problematic mct patches are still in linux-next, merged via tip tree
> by commit 9703240ef617 ("Merge branch into tip/master:
> 'timers/clocksource'"). What are the plans to resolve this?


Ouch! Thanks for letting me know. Investigating how to sort it out


-- 
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog