| Message ID | 20210322044707.19479-1-samuel@sholland.org (mailing list archive) |
|---|---|
| Headers | show |
| Series | arm64: sunxi: Enable the sun4i timer | expand |
On 22/03/2021 05:47, Samuel Holland wrote: > In preparation for adding CPU idle states, hook up the sun4i timer. > Having a non-c3stop clockevent source available is necessary for all > CPUs to simultaneously enter a local-timer-stop idle state. Why simultaneously ? > Changes from v1: > - Removed H616 changes (depends on an unmerged patch set) > - Reworded the patch 4-5 commit messages for clarity > - Added Acked-by tags > > Samuel Holland (5): > dt-bindings: timer: Simplify conditional expressions > dt-bindings: timer: Add compatibles for sun50i timers > arm64: dts: allwinner: a64: Sort watchdog node > arm64: dts: allwinner: Add sun4i MMIO timer nodes > arm64: sunxi: Build the sun4i timer driver > > .../timer/allwinner,sun4i-a10-timer.yaml | 42 +++++++++---------- > arch/arm64/Kconfig.platforms | 1 + > arch/arm64/boot/dts/allwinner/sun50i-a64.dtsi | 25 +++++++---- > arch/arm64/boot/dts/allwinner/sun50i-h6.dtsi | 9 ++++ > 4 files changed, 46 insertions(+), 31 deletions(-) >
On 3/22/21 9:18 AM, Daniel Lezcano wrote: > On 22/03/2021 05:47, Samuel Holland wrote: >> In preparation for adding CPU idle states, hook up the sun4i timer. >> Having a non-c3stop clockevent source available is necessary for all >> CPUs to simultaneously enter a local-timer-stop idle state. > > Why simultaneously ? Because the CPU providing (the hrtimer providing) the broadcast timer cannot enter an idle state which would stop that timer. So in my case, with 4 CPUs, I was seeing at most 3 CPUs enter idle at any given time. This prevented any cluster-level idle states from doing anything. After applying this series, I was able to observe the whole cluster powering down when appropriate. Regards, Samuel >> Changes from v1: >> - Removed H616 changes (depends on an unmerged patch set) >> - Reworded the patch 4-5 commit messages for clarity >> - Added Acked-by tags >> >> Samuel Holland (5): >> dt-bindings: timer: Simplify conditional expressions >> dt-bindings: timer: Add compatibles for sun50i timers >> arm64: dts: allwinner: a64: Sort watchdog node >> arm64: dts: allwinner: Add sun4i MMIO timer nodes >> arm64: sunxi: Build the sun4i timer driver >> >> .../timer/allwinner,sun4i-a10-timer.yaml | 42 +++++++++---------- >> arch/arm64/Kconfig.platforms | 1 + >> arch/arm64/boot/dts/allwinner/sun50i-a64.dtsi | 25 +++++++---- >> arch/arm64/boot/dts/allwinner/sun50i-h6.dtsi | 9 ++++ >> 4 files changed, 46 insertions(+), 31 deletions(-) >> > >
On 24/03/2021 04:51, Samuel Holland wrote: > On 3/22/21 9:18 AM, Daniel Lezcano wrote: >> On 22/03/2021 05:47, Samuel Holland wrote: >>> In preparation for adding CPU idle states, hook up the sun4i timer. >>> Having a non-c3stop clockevent source available is necessary for all >>> CPUs to simultaneously enter a local-timer-stop idle state. >> >> Why simultaneously ? > Because the CPU providing (the hrtimer providing) the broadcast timer > cannot enter an idle state which would stop that timer. So in my case, > with 4 CPUs, I was seeing at most 3 CPUs enter idle at any given time. > This prevented any cluster-level idle states from doing anything. After > applying this series, I was able to observe the whole cluster powering > down when appropriate. Ah, ok. I did not realize this fourth CPU was acting as the broadcast timer, so allowing at least a CPU power down. This setup is unusual. The changes make sense. Acked-by: Daniel Lezcano <daniel.lezcano@linaro.org>