Re: [PATCH 07/10] ARM: dts: sun7i: cubietruck: Enable the GMAC

From: Florian Fainelli <f.fainelli@gmail.com>
To: Chen-Yu Tsai <wens@csie.org>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>,
	netdev <netdev@vger.kernel.org>,
	Rob Herring <rob.herring@calxeda.com>,
	"devicetree@vger.kernel.org" <devicetree@vger.kernel.org>,
	"linux-arm-kernel@lists.infradead.org" 
	<linux-arm-kernel@lists.infradead.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	linux-sunxi <linux-sunxi@googlegroups.com>,
	Maxime Ripard <maxime.ripard@free-electrons.com>,
	Srinivas Kandagatla <srinivas.kandagatla@st.com>
Subject: Re: [PATCH 07/10] ARM: dts: sun7i: cubietruck: Enable the GMAC
Date: Mon, 9 Dec 2013 09:48:01 -0800	[thread overview]
Message-ID: <CAGVrzcaVJtoVhEfV+ZiMOkm3F8NBCf2YRcZSrZPMRfQfEr1u=g@mail.gmail.com> (raw)
In-Reply-To: <CAGb2v675Cg6S512zfM1z4b+wqE_PesDGWdgcmpzWmPzyzXqbeA@mail.gmail.com>

2013/12/8 Chen-Yu Tsai <wens@csie.org>:
> Florian, Giuseppe:
>
> On Sat, Dec 7, 2013 at 9:57 AM, Florian Fainelli <f.fainelli@gmail.com> wrote:
>> 2013/12/6 Chen-Yu Tsai <wens@csie.org>:
>>> On Sat, Dec 7, 2013 at 5:09 AM, Florian Fainelli <f.fainelli@gmail.com> wrote:
>>>> 2013/12/6 Chen-Yu Tsai <wens@csie.org>:
>>>>> The CubieTruck uses the GMAC with an RGMII phy.
>>>>>
>>>>> Signed-off-by: Chen-Yu Tsai <wens@csie.org>
>>>>> ---
>>>>>  arch/arm/boot/dts/sun7i-a20-cubietruck.dts | 8 ++++++++
>>>>>  1 file changed, 8 insertions(+)
>>>>>
>>>>> diff --git a/arch/arm/boot/dts/sun7i-a20-cubietruck.dts b/arch/arm/boot/dts/sun7i-a20-cubietruck.dts
>>>>> index 8a1009d..af212a2 100644
>>>>> --- a/arch/arm/boot/dts/sun7i-a20-cubietruck.dts
>>>>> +++ b/arch/arm/boot/dts/sun7i-a20-cubietruck.dts
>>>>> @@ -33,6 +33,14 @@
>>>>>                         pinctrl-0 = <&uart0_pins_a>;
>>>>>                         status = "okay";
>>>>>                 };
>>>>> +
>>>>> +               gmac: ethernet@01c50000 {
>>>>> +                       pinctrl-names = "default";
>>>>> +                       pinctrl-0 = <&gmac_pins_rgmii>;
>>>>> +                       snps,phy-addr = <1>;
>>>>
>>>> What is this snps,phy-addr property? Why is not a standard device tree
>>>> node for an Ethernet PHY node used?
>>>
>>> This property is implemented by stmmac and documented in the DT
>>> bindings. stmmac has not been updated to use Ethernet PHY nodes.
>>
>> This driver property should be removed and deprecated since there is
>> an ePAPR standardized Ethernet PHY node. What I am worried here is the
>> loss of information, the standard Ethernet DT node allows to specify
>> much more information (clause, maximum speed, compatible string
>> etc...).
>
> Giuseppe, any thoughts on this?
>
>>> Removing this property will not affect the function of the driver.
>>> The driver probes its MDIO bus and selects the lowest available
>>> address if not specified.
>>
>> So if this is just giving the driver a hint on where to  probe for a
>> PHY on the MDIO bus, then let's drop it and use the standard DT node
>> no?
>
> Sure. I will remove it from the DT.
>
> The stmmac driver does not have a seperate MDIO bus driver, nor
> does it support Ethernet PHY node bindings. So I will not add
> a phy node at this moment.

This will create needless churn in the DT if you do not do it now,
worse actually, we switch from a DT which specifically described
Ethernet PHY nodes properly to a version where it does, to ultimately
a newer version which does.

Considering that the absence of a "snps,phy-addr" property will still
result in the MDIO bus to be probed, keeping the existing Ethernet PHY
nodes, referencing them correclty with a "phy-handle" property, but
having no explicit support for these in the driver will not result in
a functional change, and will reduce the DT churn. Also, you could
still sneak a patch in this patchset which parses the standard
EThernet PHY node binding.

>
>>>>> +                       phy-mode = "rgmii";
>>>>
>>>> Especially since you use the relatively standard "phy-mode" property here?
>>>
>>> Used as specified in stmmac DT bindings.
>>
>> That one is kind of okay because of_get_phy_mode() knows about it,
>> even though ePAPR states this should be "phy-connection-type", so I am
>> okay with keeping that one because current practice and helpers are
>> here.
>
> It seems all ARM DTs currently use "phy-mode".

Which is why I said "is kind of okay", but this is not what the
standard came up with, ARM is not exactly the best DT-citizen out
there as people are prone to making up new properties when the
standard already defines some, anyway....
--
Florian