Re: [PATCH v3 2/2] mfd: mxs-lradc: Pass parent IRQs to child drivers

[Jonathan Cameron <jic23@kernel.org>]

On Thu, 12 Nov 2026 22:32:12 -0300
Fabio Estevam <festevam@gmail.com> wrote:

> The MFD driver stores raw DT interrupt indices in IRQ resources but
> registers its child devices without an IRQ domain or IRQ base. The child
> drivers consequently receive the indices as if they were mapped Linux
> IRQ numbers. IRQ index 0 triggers the platform_get_irq_byname() warning:
> 
>   0 is an invalid IRQ number
> 
> The children then try to recover the indices from their resources and
> map them against the parent OF node. Besides duplicating IRQ names in
> the parent and children, this relies on invalid platform IRQ resources
> and fails when IRQ 0 is rejected.
> 
> Resolve the interrupts with platform_get_irq() in the MFD parent and
> store the mapped IRQs in the shared LRADC data. Drop the child IRQ
> resources and duplicated name arrays, and have both child drivers
> request the mapped IRQs directly. Use the MFD memory base support to
> translate a single relative memory resource shared by both cells.
> 
> Tested on an i.MX28 EVK. The mxs-lradc-ts driver probes successfully,
> touch events are reported, and the touch-detect and LRADC channel 7
> interrupt counters increase as expected.
> 
> Cc: stable@vger.kernel.org
> Suggested-by: Lee Jones <lee@kernel.org>
> Closes: https://lore.kernel.org/linux-input/CAOMZO5CgTzo6GuXptD4mFTN=UsJJqpkixDW1Q64gqZGrij8+uQ@mail.gmail.com/
> Fixes: a85a6c86c25b ("driver core: platform: Clarify that IRQ 0 is invalid")
> Signed-off-by: Fabio Estevam <festevam@gmail.com>
For this change
Acked-by: Jonathan Cameron <jonathan.cameron@oss.qualcomm.com>
With assumption it'll go through mfd.

However, one of those, whilst you are here comments below....

> diff --git a/drivers/mfd/mxs-lradc.c b/drivers/mfd/mxs-lradc.c
> index 30a66e268169..d16dfea89207 100644
> --- a/drivers/mfd/mxs-lradc.c
> +++ b/drivers/mfd/mxs-lradc.c

>  static int mxs_lradc_probe(struct platform_device *pdev)
>  {
>  	struct device *dev = &pdev->dev;
>  	struct device_node *node = dev->of_node;
> +	const struct mxs_lradc_soc_data *data;
>  	struct mxs_lradc *lradc;
> -	struct mfd_cell *cells = NULL;
>  	struct resource *res;
> +	unsigned int num_cells;
>  	int ret = 0;
>  	u32 ts_wires = 0;
>  
> @@ -137,7 +161,11 @@ static int mxs_lradc_probe(struct platform_device *pdev)
>  	if (!lradc)
>  		return -ENOMEM;
>  
> -	lradc->soc = (kernel_ulong_t)device_get_match_data(&pdev->dev);
> +	data = device_get_match_data(dev);
> +	if (!data)
> +		return -ENODATA;
> +
> +	lradc->soc = data->soc;
>  
>  	lradc->clk = devm_clk_get(&pdev->dev, NULL);

This is what would need changing to fix potential ordering issues with devm / non devm
mixing.

>  	if (IS_ERR(lradc->clk)) {
> @@ -187,38 +215,15 @@ static int mxs_lradc_probe(struct platform_device *pdev)

>  
> -	ret = devm_mfd_add_devices(&pdev->dev, PLATFORM_DEVID_NONE,
> -				   &cells[TSC_CELL], 1, NULL, 0, NULL);
> +	num_cells = lradc->touchscreen_wire ? ARRAY_SIZE(mxs_lradc_cells) : 1;
> +	ret = devm_mfd_add_devices(dev, PLATFORM_DEVID_NONE,
> +				   mxs_lradc_cells, num_cells, res, 0, NULL);
>  	if (ret) {
> -		dev_err(&pdev->dev,
> -			"Failed to add the touchscreen subdevice\n");
> +		dev_err(dev, "Failed to add subdevices\n");

Not related to what you are changing here, but a devm call with a goto that
follows it is usually an alarm bell...

Can we fix that with a change to devm_clk_get_enabled()?

Jonathan


>  		goto err_clk;
>  	}