Re: [PATCH v6 14/48] mfd: ti_am335x_tscadc: Don't search the tree for our clock

[Miquel Raynal]

Hi Nikolaus,

hns@xxxxxxxxxxxxx wrote on Fri, 4 Mar 2022 23:38:25 +0100:

> Hi Miquel,
> I recently found that our BeagleBoneBlack with external touch screen stopped
> to find it.
> 
> A git bisect revealed this patch (merged into v5.16-rc1) as the first bad:
> 
> > Am 15.10.2021 um 10:14 schrieb Miquel Raynal <miquel.raynal@xxxxxxxxxxx>:
> > 
> > There is a single clock available in our node, which is named
> > "fck". The clock handler then points to adc_tsc_fck but no need to point
> > directly to it and do a full tree search.
> > 
> > Signed-off-by: Miquel Raynal <miquel.raynal@xxxxxxxxxxx>
> > ---
> > drivers/mfd/ti_am335x_tscadc.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/drivers/mfd/ti_am335x_tscadc.c b/drivers/mfd/ti_am335x_tscadc.c
> > index e2c4416f192d..8af44c08d925 100644
> > --- a/drivers/mfd/ti_am335x_tscadc.c
> > +++ b/drivers/mfd/ti_am335x_tscadc.c
> > @@ -206,7 +206,7 @@ static	int ti_tscadc_probe(struct platform_device *pdev)
> > 	 * This frequency is valid since TSC_ADC_SS controller design
> > 	 * assumes the OCP clock is at least 6x faster than the ADC clock.
> > 	 */
> > -	clk = devm_clk_get(&pdev->dev, "adc_tsc_fck");
> > +	clk = devm_clk_get(&pdev->dev, NULL);
> > 	if (IS_ERR(clk)) {
> > 		dev_err(&pdev->dev, "failed to get TSC fck\n");
> > 		err = PTR_ERR(clk);
> > -- 
> > 2.27.0
> >   
> 
> While I understand the reasons for this change there seems to be something
> missing now in the device tree because the clock isn't found any more.
> 
> After knowing about the problem I could also locate the log entry:
> 
> [    4.456680] ti_am3359-tscadc 44e0d000.tscadc: failed to get TSC fck
> 
> Reverting your patch makes it work again.

Sorry for the wrong behavior on your side and thanks for the
investigation.

> Is there missing a change in the am335x-boneblack or am335x DTS?

I've looked at the code and indeed the am33xx-clock.dtsi file defines
the touchscreen clock, but unfortunately the am33xx-l4.dtsi file which
defines the touchscreen node does not reference the clock. The bindings
clearly require the clocks to be referenced but I believe this was not
noticed until now because the clock exist and clk_get() did a lookup
across the tree.

On my side I tested it with an am437x SoC which uses another base
device tree, which properly references the touchscreen clock where it's
needed.

I will send a patch (untested), can you give it a try and report if it
fixes your issue?

> 
> Our (private) DTS looks like this, i.e. we don't play with clocks inherited
> from mainline tree:
> 
> #include "am335x-boneblack.dts"
> 
> ...
> 
> &tscadc {
>         status = "okay";
> 
>         tsc {
>                 ti,wires = <4>;
>                 ti,x-plate-resistance = <600>;
>                 ti,coordinate-readouts = <5>;
>                 ti,wire-config = <0x00 0x11 0x22 0x33>;
>         };
> 
>         adc {
>                 ti,adc-channels = <4 5 6 7>;
>         };
> };
> 
> BR and thanks,
> Nikolaus

Thanks,
Miquèl