Re: (EXT) Re: (EXT) Re: [PATCH 2/2] ARM: dts: imx6qdl: tqma6: minor fixes

From: Matthias Schiffer
Date: Tue Aug 25 2020 - 10:40:36 EST


On Tue, 2020-08-25 at 11:24 -0300, Fabio Estevam wrote:
> Hi Matthias,
>
> On Tue, Aug 25, 2020 at 4:22 AM Matthias Schiffer
> <matthias.schiffer@xxxxxxxxxxxxxxx> wrote:
>
> > Hmm, unless I'm overlooking something, this is not going to work:
> >
> > - spi_get_gpio_descs() sets num_chipselect to the maximum of the
> > num_chipselect set in the driver and the number of cs-gpios
> >
> > - spi_imx_probe() sets num_chipselect to 3 if not specified in the
> > device tree
> >
> > So I think we would end up with 3 instead of 1 chipselect.
>
> Oh, this has changed recently in 8cdcd8aeee281 ("spi: imx/fsl-lpspi:
> Convert to GPIO descriptors"):
> ....
>
> - } else {
> - u32 num_cs;
> -
> - if (!of_property_read_u32(np, "num-cs", &num_cs))
> - master->num_chipselect = num_cs;
> - /* If not preset, default value of 1 is used */
>
> Initially, if num-cs was not present the default value for
> num_chipselect was 1.
>
> - }
> + /*
> + * Get number of chip selects from device properties. This
> can be
> + * coming from device tree or boardfiles, if it is not
> defined,
> + * a default value of 3 chip selects will be used, as all the
> legacy
> + * board files have <= 3 chip selects.
> + */
> + if (!device_property_read_u32(&pdev->dev, "num-cs", &val))
> + master->num_chipselect = val;
> + else
> + master->num_chipselect = 3;
>
> Now it became 3.
>
> I think this is a driver issue and we should fix the driver instead
> of
> requiring to pass num-cs to the device tree.
>
>
> num-cs is not even documented in the spi-imx binding.

Makes sense. Does the following logic sound correct?

- If num-cs is set, use that (and add it to the docs)
- If num-cs is unset, use the number of cs-gpios
- If num-cs is unset and no cs-gpios are defined, use a driver-provided
default


I'm not sure if 3 is a particularly useful default either, but it seems
it was chosen to accommodate boards that previously set this via
platform data. All SoCs I've checked (i.MX6Q/DL, i.MX6UL, i.MX7) have 4
internal CS pins per ECSPI instance, so maybe the driver should use
that as its default instead?