Re: [PATCH] ARM: EXYNOS: Clear OF_POPULATED flag from PMU node in IRQ init callback

[Krzysztof Kozlowski]

On Sun, Aug 21, 2016 at 03:27:45AM -0400, Javier Martinez Canillas wrote:
> The Exynos PMU node is an interrupt, clock and PMU (Power Management Unit)
> controller, and these functionalities are supported by different drivers
> that matches the same compatible strings.
> 
> Since commit 15cc2ed6dcf9 ("of/irq: Mark initialised interrupt controllers
> as populated") the OF core flags interrupt controllers registered with the
> IRQCHIP_DECLARE() macro as OF_POPULATED, so platform devices with the same
> compatible string as the interrupt controller will not be registered.
> 
> This prevents the PMU platform device to be registered so the Exynos PMU
> driver is never probed. This breaks (among other things) Suspend-to-RAM.
> 
> Fix this by clearing the OF_POPULATED flag in the PMU IRQ init callback,
> to allow the Exynos PMU platform driver to be probed. The patch is based
> on Philipp Zabel's "ARM: imx6: mark GPC node as not populated after irq
> init to probe pm domain driver".
> 
> Fixes: 15cc2ed6dcf9 ("of/irq: Mark initialised interrupt controllers as populated")
> Signed-off-by: Javier Martinez Canillas <javier@xxxxxxxxxxxxxxx>
> 
> ---
> 
> This was tested on an Exynos5800 Peach Pi Chromebook with v4.8-rc2 +
> Philipp's patch: https://patchwork.kernel.org/patch/9271361/
> 
>  arch/arm/mach-exynos/suspend.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/arch/arm/mach-exynos/suspend.c b/arch/arm/mach-exynos/suspend.c
> index 3750575c73c5..06332f626565 100644
> --- a/arch/arm/mach-exynos/suspend.c
> +++ b/arch/arm/mach-exynos/suspend.c
> @@ -255,6 +255,12 @@ static int __init exynos_pmu_irq_init(struct device_node *node,
>  		return -ENOMEM;
>  	}
>  
> +	/*
> +	 * Clear the OF_POPULATED flag set in of_irq_init so that
> +	 * later the Exynos PMU platform device won't be skipped.
> +	 */
> +	of_node_clear_flag(node, OF_POPULATED);
> +

Looks like a proper workaround (at least till of_irq_init sets this flag
before calling irq_init_cb()) however I have some more doubts:

1. The commit 15cc2ed6dcf9 ("of/irq: Mark initialised interrupt
controllers as populated") might be also applied to clock CLK_OF_DECLARE
and the problem will appear again.

2. We are reusing PMU compatible for irqcip, clock provider and a
platform driver. This is one PMU block with many features thus many
drivers were created. What happens if the Exynos platform driver
(drivers/soc/samsung/exynos-pmu.c) binds before irqchip or clk provider?
Probably we should not reuse the compatible but create a new one for
each type of driver? How does it match DeviceTree?

Javier, Rob, any hints on that? Best practices, suggestions for future?

Best regards,
Krzysztof