Re: [PATCH v5 6/7] dt-bindings: pinctrl: mediatek,mt6779-pinctrl: Document MT6735 pin controller
From: Yassine Oudjana
Date: Mon Nov 21 2022 - 11:01:43 EST
On Mon, Nov 21 2022 at 14:48:58 +01:00:00, AngeloGioacchino Del Regno
<angelogioacchino.delregno@xxxxxxxxxxxxx> wrote:
Il 18/11/22 12:30, Yassine Oudjana ha scritto:
From: Yassine Oudjana <y.oudjana@xxxxxxxxxxxxxx>
Add bindings for the pin controller found on MediaTek MT6735 and
MT6735M SoCs, including describing a method to manually specify
a pin and function in the pinmux property making defining bindings
for each pin/function combination unnecessary. The pin controllers
on those SoCs are generally identical, with the only difference
being the lack of MSDC2 pins (198-203) on MT6735M.
Signed-off-by: Yassine Oudjana <y.oudjana@xxxxxxxxxxxxxx>
Reviewed-by: Rob Herring <robh@xxxxxxxxxx>
---
.../pinctrl/mediatek,mt6779-pinctrl.yaml | 55
++++++++++++++++++-
MAINTAINERS | 6 ++
2 files changed, 60 insertions(+), 1 deletion(-)
..snip..
@@ -352,18 +391,32 @@ examples:
};
/* GPIO0 set as multifunction GPIO0 */
- gpio-pins {
+ gpio0-pins {
pins {
pinmux = <PINMUX_GPIO0__FUNC_GPIO0>;
};
};
+ /* GPIO1 set to function 0 (GPIO) */
+ gpio1-pins {
+ pins {
+ pinmux = <(MTK_PIN_NO(1) | 0)>;
Please follow the same format that you can find in all of the
mtXXXX-pinfunc.h.
What you wrote here (MTK_PIN_NO(x) | func) is defined in there for
the purpose
of providing a definition name that actually means something (for
both readability
and documentation purposes).
This means that your GPIO1 set to function 0 (gpio) should be
pinmux = <PINMUX_GPIO1__FUNC_GPIO1>;
+ };
+ };
+
/* GPIO52 set as multifunction SDA0 */
i2c0-pins {
pins {
pinmux = <PINMUX_GPIO52__FUNC_SDA0>;
};
};
+
+ /* GPIO62 set to function 1 (primary function) */
+ i2c1-pins {
+ pins {
+ pinmux = <(MTK_PIN_NO(62) | 1)>;
pinmux = <PINMUX_GPIO62__FUNC_SDA1>; (is it sda1??)
This means that you should as well add a mediatek,mt6735-pinfunc.h
binding...
This is pretty much what I was trying to avoid by doing this.
Originally I tried to have something similar to qualcomm pin
controllers which use "pins" and "function" properties (but probably
with integer values rather than strings) without making any major
changes to pinctrl-paris or related DT bindings, but it quickly became
obvious that it wouldn't be possible when looking at how it does things
currently. pinctrl-moore was better in this aspect, actually making use
of pin groups to describe how sets of pins have shared functions
instead of making a group for each pin, and taking "groups" and
"function" properties. However, it wasn't fully suitable for the
hardware so I had to stick with pinctrl-paris. At that point I thought
of this to be the simplest way of doing it. I think it is unnecessary
to define every single pin-function combination. Yes, doing it this way
doesn't make it clear what function is being set right away, but a
quick look at pinctrl-mtk-mt6735.h is all it takes to find out.
Furthermore, in most cases functions 0 (GPIO) and 1 (primary, pin named
after it) are the only ones used so knowing the pin names is all it
takes to figure out the functions.
With all of that being said however, I guess I don't mind following the
current convention for the time being. The pinctrl subsystem (not just
mediatek pin controllers) has some of the most inconsistent DT bindings
from what I've seen, specifically when it comes to specifying pin
functions, and I think it will end up having some major cleanup down
the line anyway.