Received-SPF: pass (google.com: domain of linux-kernel+bounces-184884-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
Precedence: bulk
MIME-Version: 1.0
References: <20240519144920.14804-1-egyszeregy@freemail.hu>
 <CAMRc=MctEsMiRgaV5UTiaxjoFDa2izX9wnLAU07=G8gBEcSKoQ@mail.gmail.com> <3c0f28a0-9d07-4e15-ad8e-ad22f25ff979@freemail.hu>
In-Reply-To: <3c0f28a0-9d07-4e15-ad8e-ad22f25ff979@freemail.hu>
From: Bartosz Golaszewski <brgl@bgdev.pl>
Date: Tue, 21 May 2024 13:36:24 +0200
Message-ID: <CAMRc=MeiDJoZe+X8CzGduFdSs4dTo8L3FPPd30LW07VpH_wG7A@mail.gmail.com>
Subject: Re: [PATCH] gpiolib: Introduce "linux,gpiochip-name" property for
 device tree of GPIO controller.
To: =?UTF-8?Q?Sz=C5=91ke_Benjamin?= <egyszeregy@freemail.hu>
Cc: linus.walleij@linaro.org, linux-gpio@vger.kernel.org, 
	linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Mon, May 20, 2024 at 6:41=E2=80=AFPM Sz=C5=91ke Benjamin <egyszeregy@fre=
email.hu> wrote:
>
> 2024. 05. 19. 19:27 keltez=C3=A9ssel, Bartosz Golaszewski =C3=ADrta:
> > On Sun, May 19, 2024 at 4:49=E2=80=AFPM <egyszeregy@freemail.hu> wrote:
> >>
> >> From: Benjamin Sz=C5=91ke <egyszeregy@freemail.hu>
> >>
> >> Optionally, a GPIO controller may have a "linux,gpiochip-name" propert=
y.
> >
> > Oh, may it really?
> >
> > $ git grep "linux,gpiochip-name" Documentation/devicetree/bindings/
> > $
> >
> > Doesn't look like it.
> >
> >> This is a string which is defining a custom suffix name for gpiochip i=
n
> >> /dev/gpiochip-<name> format. It helps to improve software portability
> >> between various SoCs and reduce complexities of hardware related codes
> >> in SWs.
> >>
> >> Signed-off-by: Benjamin Sz=C5=91ke <egyszeregy@freemail.hu>
> >> ---
> >>   drivers/gpio/gpiolib.c | 12 +++++++++++-
> >>   1 file changed, 11 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
> >> index ce94e37bcbee..e24d8db1d054 100644
> >> --- a/drivers/gpio/gpiolib.c
> >> +++ b/drivers/gpio/gpiolib.c
> >> @@ -860,6 +860,7 @@ int gpiochip_add_data_with_key(struct gpio_chip *g=
c, void *data,
> >>                                 struct lock_class_key *lock_key,
> >>                                 struct lock_class_key *request_key)
> >>   {
> >> +       const char *name;
> >>          struct gpio_device *gdev;
> >>          unsigned int desc_index;
> >>          int base =3D 0;
> >> @@ -896,7 +897,16 @@ int gpiochip_add_data_with_key(struct gpio_chip *=
gc, void *data,
> >>                  goto err_free_gdev;
> >>          }
> >>
> >> -       ret =3D dev_set_name(&gdev->dev, GPIOCHIP_NAME "%d", gdev->id)=
;
> >> +       /*
> >> +        * If "linux,gpiochip-name" is specified in device tree, use /=
dev/gpiochip-<name>
> >> +        * in Linux userspace, otherwise use /dev/gpiochip<id>.
> >> +        */
> >> +       ret =3D device_property_read_string(gc->parent, "linux,gpiochi=
p-name", &name);
> >> +       if (ret < 0)
> >> +               ret =3D dev_set_name(&gdev->dev, GPIOCHIP_NAME "%d", g=
dev->id);
> >> +       else
> >> +               ret =3D dev_set_name(&gdev->dev, GPIOCHIP_NAME "-%s", =
name);
> >> +
> >>          if (ret)
> >>                  goto err_free_ida;
> >>
> >> --
> >> 2.39.3
> >>
> >
> > NAK, this is udev's task in user-space if you need this.
> >
> > Bart
>
> Hi,
>
> Goal of this patch is to introduce this new mode to assign a custom name =
from
> lowlevel device tree to a gpio bank.
>

The goal of this patch seems to be to push some unnecessary downstream
functionality into the mainline kernel.

> It is more maintainable then use udev in userspace for it.
> For example there are three different SoCs: i.MX7, i.MX9, ZynqMP.
>
> In Yocto project, the Linux image's SW environment is nicely configurable
> independently from what is the target MACHNIE. But if i like to deploy a =
SW
> which uses peripheries like gpiobanks, i2c-dev, spidev these /dev/... nam=
e will
> be totally different on each SoCs, more over in ZynqMP and any other Adap=
tive
> SoC platform, the index number for the gpiobanks or other can be not
> deterministic if it probed in run-time. Goal is to easily make a Linux OS=
 image
> which can support multiple SoCs and in SW point of view it is flexible.
>
> So, in Yocto project point of view the best, if any Machine specific sett=
ings is
> stored in the device tree files of the target MACHNIEs in driver levels/c=
onfig,
> because it will be deterministic in 100% sure and it will be nicely separ=
ated
> from the SW meta layers which may not contains any machine specific hacki=
ng with
> udev and so on for building image.
>
> So this way to assign a custom name for a gpiobanks from device tree is m=
ore
> efficient and more maintainable in SW developing point of view in
> Yocto/buildroot world because i need to just define a name like
> linux,gpiochip-name =3D "expander"; then use it with a fixed name in my g=
eneric SW
> under /dev/gpiochip-expander name. And there are no need to care about wh=
at will
> be the index number of this banks randomly after boot and how need to mak=
e an
> ugly append layer for udev and make it for all of machine variants which =
are the
> target to use with the same SW.
>
> My opinion udev is ugly to use for it and more complicated, and no longer
> beneficial for new Adaptive SoCs where they can be not deterministic what=
 kind
> of index number they got in driver probing (you will not know what index =
need to
> mapping for what custom name). It is much better to assign this suffix/na=
me
> explicitly from device tree.
>

The information you want to store here is already accessible in the
chip's label (set by the relevant driver) that you can retrieve using
libgpiod (or sysfs if you must) and create appropriate links using
udev. This is how it's done for other devices, I see no reason to make
GPIO a special case.

Bart