[5/9] arm64: dts: qcom: sa8775p-ride: move the reset-gpios property of the PHY

  From: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>

Device-tree bindings for MDIO define per-PHY reset-gpios as well as a
global reset-gpios property at the MDIO node level which controlls all
devices on the bus. The latter is most likely a workaround for the
chicken-and-egg problem where we cannot read the ID of the PHY before
bringing it out of reset but we cannot bring it out of reset until we've
read its ID.

I have proposed a solution for this problem in 2020 but it never got
upstream. Now we have a workaround in place which allows us to hard-code
the PHY id in the compatible property, thus skipping the ID scanning).

Let's make the device-tree for sa8775p-ride slightly more correct by
moving the reset-gpios property to the PHY node with its ID put into the
PHY node's compatible.

Link: https://lore.kernel.org/all/20200622093744.13685-1-brgl@bgdev.pl/
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>
---
 arch/arm64/boot/dts/qcom/sa8775p-ride.dts | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)
  

On Mon, Aug 07, 2023 at 09:35:03PM +0200, Bartosz Golaszewski wrote:
> From: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>
> 
> Device-tree bindings for MDIO define per-PHY reset-gpios as well as a
> global reset-gpios property at the MDIO node level which controlls all

s/controlls/controls/

> devices on the bus. The latter is most likely a workaround for the
> chicken-and-egg problem where we cannot read the ID of the PHY before
> bringing it out of reset but we cannot bring it out of reset until we've
> read its ID.
> 
> I have proposed a solution for this problem in 2020 but it never got
> upstream. Now we have a workaround in place which allows us to hard-code
> the PHY id in the compatible property, thus skipping the ID scanning).

nitpicky, but I think that already existed at that time :D

> 
> Let's make the device-tree for sa8775p-ride slightly more correct by
> moving the reset-gpios property to the PHY node with its ID put into the
> PHY node's compatible.
> 
> Link: https://lore.kernel.org/all/20200622093744.13685-1-brgl@bgdev.pl/
> Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>
> ---
>  arch/arm64/boot/dts/qcom/sa8775p-ride.dts | 7 +++----
>  1 file changed, 3 insertions(+), 4 deletions(-)
> 
> diff --git a/arch/arm64/boot/dts/qcom/sa8775p-ride.dts b/arch/arm64/boot/dts/qcom/sa8775p-ride.dts
> index 38327aff18b0..1c471278d441 100644
> --- a/arch/arm64/boot/dts/qcom/sa8775p-ride.dts
> +++ b/arch/arm64/boot/dts/qcom/sa8775p-ride.dts
> @@ -279,13 +279,12 @@ mdio {
>  		#address-cells = <1>;
>  		#size-cells = <0>;
>  
> -		reset-gpios = <&pmm8654au_2_gpios 8 GPIO_ACTIVE_LOW>;
> -		reset-delay-us = <11000>;
> -		reset-post-delay-us = <70000>;
> -
>  		sgmii_phy: phy@8 {
> +			compatible = "ethernet-phy-id0141.0dd4";
>  			reg = <0x8>;
>  			device_type = "ethernet-phy";
> +			reset-gpios = <&pmm8654au_2_gpios 8 GPIO_ACTIVE_LOW>;
> +			reset-deassert-us = <70000>;

Doesn't this need reset-assert-us?
  

> > I have proposed a solution for this problem in 2020 but it never got
> > upstream. Now we have a workaround in place which allows us to hard-code
> > the PHY id in the compatible property, thus skipping the ID scanning).
> 
> nitpicky, but I think that already existed at that time :D

Yes, it has been there are long long time. It is however only in the
last 5 years of so has it been seen as a solution to the chicken egg
problem.

> >  		sgmii_phy: phy@8 {
> > +			compatible = "ethernet-phy-id0141.0dd4";
> >  			reg = <0x8>;
> >  			device_type = "ethernet-phy";
> > +			reset-gpios = <&pmm8654au_2_gpios 8 GPIO_ACTIVE_LOW>;
> > +			reset-deassert-us = <70000>;
> 
> Doesn't this need reset-assert-us?

If i remember correctly, there is a default value if DT does not
provide one.

	Andrew
  

On Mon, Aug 07, 2023 at 11:51:40PM +0200, Andrew Lunn wrote:
> > > I have proposed a solution for this problem in 2020 but it never got
> > > upstream. Now we have a workaround in place which allows us to hard-code
> > > the PHY id in the compatible property, thus skipping the ID scanning).
> > 
> > nitpicky, but I think that already existed at that time :D
> 
> Yes, it has been there are long long time. It is however only in the
> last 5 years of so has it been seen as a solution to the chicken egg
> problem.
> 
> > >  		sgmii_phy: phy@8 {
> > > +			compatible = "ethernet-phy-id0141.0dd4";
> > >  			reg = <0x8>;
> > >  			device_type = "ethernet-phy";
> > > +			reset-gpios = <&pmm8654au_2_gpios 8 GPIO_ACTIVE_LOW>;
> > > +			reset-deassert-us = <70000>;
> > 
> > Doesn't this need reset-assert-us?
> 
> If i remember correctly, there is a default value if DT does not
> provide one.
> 

I've been trying to make sure I view devicetree properties as an OS
agnostic ABI lately, with that in mind...

The dt-binding says this for ethernet-phy:

  reset-assert-us:
    description:
      Delay after the reset was asserted in microseconds. If this
      property is missing the delay will be skipped.

If the hardware needs a delay I think we should encode it based on that
description, else we risk it starting to look like a unit impulse!
  

On Tue, Aug 8, 2023 at 12:27 AM Andrew Halaney <ahalaney@redhat.com> wrote:
>
> On Mon, Aug 07, 2023 at 11:51:40PM +0200, Andrew Lunn wrote:
> > > > I have proposed a solution for this problem in 2020 but it never got
> > > > upstream. Now we have a workaround in place which allows us to hard-code
> > > > the PHY id in the compatible property, thus skipping the ID scanning).
> > >
> > > nitpicky, but I think that already existed at that time :D
> >
> > Yes, it has been there are long long time. It is however only in the
> > last 5 years of so has it been seen as a solution to the chicken egg
> > problem.
> >
> > > >           sgmii_phy: phy@8 {
> > > > +                 compatible = "ethernet-phy-id0141.0dd4";
> > > >                   reg = <0x8>;
> > > >                   device_type = "ethernet-phy";
> > > > +                 reset-gpios = <&pmm8654au_2_gpios 8 GPIO_ACTIVE_LOW>;
> > > > +                 reset-deassert-us = <70000>;
> > >
> > > Doesn't this need reset-assert-us?
> >
> > If i remember correctly, there is a default value if DT does not
> > provide one.
> >
>
> I've been trying to make sure I view devicetree properties as an OS
> agnostic ABI lately, with that in mind...
>
> The dt-binding says this for ethernet-phy:
>
>   reset-assert-us:
>     description:
>       Delay after the reset was asserted in microseconds. If this
>       property is missing the delay will be skipped.
>
> If the hardware needs a delay I think we should encode it based on that
> description, else we risk it starting to look like a unit impulse!
>

Please note that the mdio-level delay properties are not the same as
the ones on the PHY levels.

reset-delay-us - this is the delay BEFORE *DEASSERTING* the reset line
reset-post-delay-us - this is the delay AFTER *DEASSERTING* the reset line

On PHY level we have:

reset-assert-us - AFTER *ASSERTING*
reset-deassert-us - AFTER *DEASSERTING*

There never has been any reset-assert delay on that line before. It
doesn't look like we need a delay BEFORE deasserting the line, do we?

Bart
  

> I've been trying to make sure I view devicetree properties as an OS
> agnostic ABI lately, with that in mind...
> 
> The dt-binding says this for ethernet-phy:
> 
>   reset-assert-us:
>     description:
>       Delay after the reset was asserted in microseconds. If this
>       property is missing the delay will be skipped.
> 
> If the hardware needs a delay I think we should encode it based on that
> description, else we risk it starting to look like a unit impulse!
 
I checked, and the documentation does appear correct, there is no
default value. So yes, it does seem prudent to specify a value,
otherwise it could be a short pulse.

	  Andrew
  

On Tue, Aug 08, 2023 at 02:16:50PM +0200, Bartosz Golaszewski wrote:
> On Tue, Aug 8, 2023 at 12:27 AM Andrew Halaney <ahalaney@redhat.com> wrote:
> >
> > On Mon, Aug 07, 2023 at 11:51:40PM +0200, Andrew Lunn wrote:
> > > > > I have proposed a solution for this problem in 2020 but it never got
> > > > > upstream. Now we have a workaround in place which allows us to hard-code
> > > > > the PHY id in the compatible property, thus skipping the ID scanning).
> > > >
> > > > nitpicky, but I think that already existed at that time :D
> > >
> > > Yes, it has been there are long long time. It is however only in the
> > > last 5 years of so has it been seen as a solution to the chicken egg
> > > problem.
> > >
> > > > >           sgmii_phy: phy@8 {
> > > > > +                 compatible = "ethernet-phy-id0141.0dd4";
> > > > >                   reg = <0x8>;
> > > > >                   device_type = "ethernet-phy";
> > > > > +                 reset-gpios = <&pmm8654au_2_gpios 8 GPIO_ACTIVE_LOW>;
> > > > > +                 reset-deassert-us = <70000>;
> > > >
> > > > Doesn't this need reset-assert-us?
> > >
> > > If i remember correctly, there is a default value if DT does not
> > > provide one.
> > >
> >
> > I've been trying to make sure I view devicetree properties as an OS
> > agnostic ABI lately, with that in mind...
> >
> > The dt-binding says this for ethernet-phy:
> >
> >   reset-assert-us:
> >     description:
> >       Delay after the reset was asserted in microseconds. If this
> >       property is missing the delay will be skipped.
> >
> > If the hardware needs a delay I think we should encode it based on that
> > description, else we risk it starting to look like a unit impulse!
> >
> 
> Please note that the mdio-level delay properties are not the same as
> the ones on the PHY levels.
> 
> reset-delay-us - this is the delay BEFORE *DEASSERTING* the reset line
> reset-post-delay-us - this is the delay AFTER *DEASSERTING* the reset line
> 
> On PHY level we have:
> 
> reset-assert-us - AFTER *ASSERTING*
> reset-deassert-us - AFTER *DEASSERTING*
> 
> There never has been any reset-assert delay on that line before. It
> doesn't look like we need a delay BEFORE deasserting the line, do we?
> 

The MDIO reset-delay-us happens "before deassert"/"after assert", so to
make things a proper move here I think it needs a resrt-assert, otherwise
behavior with respect to reset timing is definitely changed from this
patch!

Here's a trimmed version of the reset handling in mdio_bus.c to back
that up:

	/* assert bus level PHY GPIO reset */
	gpiod = devm_gpiod_get_optional(&bus->dev, "reset", GPIOD_OUT_HIGH);
    ...
	} else	if (gpiod) {
		bus->reset_gpiod = gpiod;
		fsleep(bus->reset_delay_us);
		gpiod_set_value_cansleep(gpiod, 0);
		if (bus->reset_post_delay_us > 0)
			fsleep(bus->reset_post_delay_us);
	}

so its assert reset, sleep reset_delay_us, deassert, sleep
reset_post_delay_us for the MDIO case.

Thanks,
Andrew