On 28/03/2019 14:33, Marek Szyprowski wrote:
> Since commit 45f1ff59e27c ("cpuidle: Return nohz hint from
> cpuidle_select()") Exynos CPUidle driver stopped entering C1 (AFTR) mode
> on Exynos4412-based Trats2 board.
>
> Further analysis revealed that the CPUidle framework changed the way
> it handles predicted timer ticks and reported target residency for the
> given idle states. As a result, the C1 (AFTR) state was not chosen
> anymore on completely idle device. The main issue was to high target
> residency value. The similar C1 (AFTR) state for 'coupled' CPUidle
> version used 10 times lower value for the target residency, despite
> the fact that it is the same state from the hardware perspective.
>
> The 100000us value for standard C1 (AFTR) mode is there from the begining
> of the support for this idle state, added by the commit 67173ca492ab
> ("ARM: EXYNOS: Add support AFTR mode on EXYNOS4210"). That commit doesn't
> give any reason for it, instead it looks like it was blindly copied from
> the WFI/IDLE state of the same driver that time. That time, that value
> was probably not really used by the framework for any critical decision,
> so it didn't matter that much.
>
> Now it turned out to be an issue, so unify the target residency with the
> 'coupled' version, as it seems to better match the real use case values
> and restores the operation of the Exynos CPUidle driver on the idle
> device.
>
> Signed-off-by: Marek Szyprowski <m.szyprowski@xxxxxxxxxxx>
> ---
Acked-by: Daniel Lezcano <daniel.lezcano@xxxxxxxxxx>
--
<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
