+ Jon

[...]

quoted

In regards to model the HW correctly. I don't believe your HW topology have
each PWM block, GPIO block, etc, in their own gate-able power domain, right?
Or is it so?

First of all, thanks for your and Rob's good suggestions.
I spent a lot more time to do research to try to make it better according to your
suggestions. Some of your ideas absolutely helped me a lot and gave me some
inspirations.

Some new findings are that some of our power domains are organized in a tree
was because the kernel does not support multi-power domains for one device.
but from HW point of view, they're identical.
You can more serious situation from local tree at here:
https://source.codeaurora.org/external/imx/linux-imx/tree/arch/arm64/boot/dts/freescale/fsl-imx8dx.dtsi?h=imx_4.9.88_imx8qxp_beta2#n195

I just found your following patch perfectly addressed our issue as we do have many
Devices requires to handle multi power domains.
3c095f32a92b PM / Domains: Add support for multi PM domains per device to genpd
e.g. For LVDS devices, it need both LVDS and DC(Display controller) PD to work.

Thanks for letting me know about further use cases about multiple PM
domain per device. Very interesting.

Now i was trying to understand all of our current user case with the team, If all of them
do not need a tree to address dependency issue, I may remove them all from device tree.
Then it would be much more clearer than before and be something like what Tegra does.
include/dt-bindings/power/tegra186-powergate.h
drivers/soc/tegra/powergate-bpmp.c

Yes!

For example, 1 #power-domain-cells will be enough. The first cell can be a global SCU
power domain and the 2nd cell the device ID.
Hopefully this can be what you and Rob want.

If any good suggestions please let me know.

It makes perfect sense!

In regards to deploy support for multiple PM domains per device, you
may have a look at the recent work done Jon Hunter for the Tegra xHCI
genpd support [1].

[...]

Kind regards
Uffe

[1]
https://patchwork.ozlabs.org/cover/976328/