Re: [PATCH RFT/RFC] mfd: cs42l43: setup true links with software nodes

[Bartosz Golaszewski]

On Wed, Nov 19, 2025 at 10:58 AM Charles Keepax
<ckeepax@xxxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Wed, Nov 19, 2025 at 10:40:30AM +0100, Bartosz Golaszewski wrote:
> > On Wed, Nov 19, 2025 at 10:31 AM Charles Keepax
> > <ckeepax@xxxxxxxxxxxxxxxxxxxxx> wrote:
> > >
> > > On Wed, Nov 19, 2025 at 10:10:24AM +0100, Bartosz Golaszewski wrote:
> > > > From: Bartosz Golaszewski <bartosz.golaszewski@xxxxxxxxxx>
> > > >
> > > > Currently the SPI driver sets up a software node rerefencing the GPIO
> > > > controller exposing the chip-select GPIO but this node never gets
> > > > attached to the actual GPIO provider. The lookup uses the weird way GPIO
> > > > core performs the software node lookup by the swnode's name. We want to
> > > > switch to a true firmware node lookup so the actual link must exist.
> > > > Move the configuration to the MFD core and connect the SPI and pinctrl
> > > > sub-devices with software node references.
> > > >
> > > > Fixes: 439fbc97502a ("spi: cs42l43: Add bridged cs35l56 amplifiers")
> > > > Reported-by: Charles Keepax <ckeepax@xxxxxxxxxxxxxxxxxxxxx>
> > > > Closes: https://lore.kernel.org/all/aRyf7qDdHKABppP8@xxxxxxxxxxxxxxxxxxxxx/
> > > > Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@xxxxxxxxxx>
> > > > ---
> > > > Hi Charles!
> > > >
> > > > Please give this a try. It's only build-tested so far. I hope I
> > > > understood correctly that it's the SPI driver that needs the "cs" GPIO
> > > > from the pinctrl.
> > >
> > > Unfortunately it fails with an -EBUSY on adding the MFD children.
> > > I will investigate a little more and report back.
> >
> > Does it fail in device_add_software_node() in MFD core? Is it possible
> > that the device already has a software node for some reason?
>
> Yeah it seems there is already a software node, although somewhat
> at a loss as to why, the only ones we add are inside the SPI
> driver. I will poke further and see if I can find out how it
> acquires that node.
>
> > > > +static const struct software_node cs42l43_gpiochip_swnode = {
> > > > +     .name = "cs42l43-pinctrl",
> > > > +};
> > > > +
> > > > +static const struct property_entry cs42l43_cs_props[] = {
> > > > +     PROPERTY_ENTRY_GPIO("cs-gpios", &cs42l43_gpiochip_swnode, 0, GPIO_ACTIVE_LOW),
> > > > +     { }
> > > > +};
> > >
> > > This drops the undefined node, that is important as it lets the
> > > SPI core know that device is using an internal chip select rather
> > > than a GPIO.
> >
> > I really dislike the whole concept of this undefined software node.
> > This driver is literally the only user and I'd love to just get rid of
> > it. HOW exactly does it let the driver know to use internal chip
> > select? Can we do it differently?
>
> The SPI code lives in spi_get_gpio_descs(). The basic gist is
> it will use a native chip select if the gpiod_get_index return
> NULL. The system came from device tree (see Documentation
> spi-controller.yaml) initially, then I added an analogue to
> swnodes a while back in 9d50f95bc0d5 ("gpio: swnode: Add
> ability to specify native chip selects for SPI").  I have no
> great attachment to this way of doing it, however, it does
> seem logical to me that the system is at least fairly similar
> between the different firmware types.
>

I see. Then maybe we should at least generalize it to a high-level
"swnode-undefined", move it out of GPIOLIB and provide a proper swnode
interface "swnode_is_undefined()" so that we don't have to open-code
the string comparison like that. That's not related to this patch of
course, just throwing out an idea.

Bart