Re: [PATCH v2 1/9] fixed-phy: register fixed PHY as platform driver

2013/12/19 Hauke Mehrtens [off-list ref]:

On 12/19/2013 02:48 AM, Florian Fainelli wrote:

quoted

2013/12/18 Hauke Mehrtens [off-list ref]:

quoted

This changes the fixed phy driver from registering the mdio bus when
the module gets loaded to registering it when a device was registered.
A phy has to get registered to this driver before it registered the
mdio bus, but this only worked when the phys are registered in some
arch code before the system booted completely. Now we want to do so
when the Ethernet driver gets initialized which could be happen every
time.

To make this driver work with such a case, convert it to a platform
driver which could be registered every time with the phys which should
be on the bus.

This was only tested on BCM47XX, but not on AR7 because I do not have
such a device.

I do understand why you would want to do it that way, but I believe
this is should be addressed separately, outside of the actual b44
changes. Converting the fixed PHY driver into a platform driver also
has an impact on how Device Tree callers such as PowerPC would be
dealing with this.

For the ADM switch we could initialize the fixed phy in the arch code,
because with that switch fixed phys are always needed, but this is also
needed for some Broadcom switches and I can not automatically detect if
that is the case. The detection could be done  based on the board the
kernel was booted on, but there are some hundreds different ones for
BCM47XX.

Cannot we unconditionally register a fixed PHY at address 0 and use it whether:

- we are connected to an ADM switch
- we have a dual-MAC configuration

It seems to me like this would always work, the driver is the one
deciding which MDIO bus to bind the PHY to.

Sorry I missed some calls to fixed_phy_ads() in the arch code where it
is used by the device tree platforms, but that could be easily converted
to a platform device as well.

quoted

Can you submit the required changes to arch/mips/bcm47xx/ for now and
get this change merged via David's tree? This would buy us some time
to discuss how to best deal with fixed PHY, and also take Device Tree
aware platform into account since that part has been an on-going
discussion for a while.

What changes to arch/mips/bcm47xx/ are you talking about?

I am talking about the call to fixed_phy_add() which have to happen
early enough before the fixed MDIO bus is probed, just like what AR7
does currently.

I do agree though that something needs to be done because the fixed
MDIO bus usage comes from days where it was easy to add a call to
fixed_phy_add() early enough in your platform code, this is no longer
the case with Device Tree and such.

Do you mean I should send the b44 patches without the fixed phy stuff
first and then we can discuss the fixed phy stuff separately, that would
be my next strategy? ;-)

--
Florian