Re: [PATCH] net: davinci_mdio: add GPIO reset logic

[Florian Fainelli <f.fainelli@gmail.com>]

On 04/08/2017 08:10 AM, Andrew Lunn wrote:
> On Sat, Apr 08, 2017 at 06:55:45AM -0700, David Miller wrote:
>> From: Roger Quadros <rogerq@ti.com>
>> Date: Wed, 5 Apr 2017 11:33:57 +0300
>>
>>> Some boards [1] leave the PHYs at an invalid state
>>> during system power-up or reset thus causing unreliability
>>> issues with the PHY like not being detected by the mdio bus
>>> or link not functional. To work around these boards have
>>> a GPIO connected to the PHY's reset pin.
>>>
>>> Implement GPIO reset handling for such cases.
>>>
>>> [1] - am572x-idk, am571x-idk, a437x-idk.
>>>
>>> Signed-off-by: Roger Quadros <rogerq@ti.com>
>>> Signed-off-by: Sekhar Nori <nsekhar@ti.com>
>>
>> I have not seen a resolution in this discussion.
>>
>> My understanding is that there are several cases (single MDIO bus whose
>> reset does a reset on all that MDIO bus's PHYs, etc.) and it's unclear
>> how to handle all such cases cleanly.
> 
> I see it falling into two cases.
> 
> 1) We have a GPIO which resets one PHY. In this case, the GPIO is a
> PHY property, it should be documented in
> Documentation/devicetree/bindings/net/phy.txt. Hopefully there is
> nothing PHY driver specific here, so all the handling can be placed in
> the core PHY code.

I suspect we would have to release the PHY GPIO reset line within a
mii_bus::reset callback, which occurs before the PHY registers are read.
There is this chicken and egg problem where you can't probe for a PHY
unless you can successfully read from it, and you can't do the PHY reset
in the PHY drivers' probe function unless you were able to find a PHY
device.

NB: you can work around that by using an Ethernet PHY device compatible
string in Device Tree that already has the PHY OUI specified, although
that usually does not scale to many different boards/designs.

> 
> 2) We have one or more GPIOs which reset more than one PHY. In this
> case, the GPIOs are MDIO bus properties. Again, there should not be
> anything which is MDIO bus driver specific, so all the handling can be
> placed in the core MDIO bus code.

Agreed.

I would do something like:

- if the MDIO bus node has a "gpio" reset property, use it and release
the device from reset
- for each available child node:
	- if the PHY/MDIO device's node have a "gpio" reset property use it and
release the PHYs from reset.

All of this should probably be placed somewhere in drivers/of/of_mdio.c
and deal with conditional GPIO/RESET controller support.
-- 
Florian