On 11/14/18 1:16 PM, Andrew Lunn wrote:
>>> Thats again because of this product has tightly integrated MAC+Phy.
>>> MAC FW controls system interface and reports/alters link state
>>> as a joint state on copper and SIF (even in dpa direct phy mode).
>>>
>>> We can't extract phy api into a standalone fully functional phylib therefore.
>>> Also as far as I know this particular phy is not available in the wild.
>>
>> So the point is that MAC firmware is managing SERDES and system interface link.
>
> Linux can manage that SERDES link between the MAC and the PHY. There
> are two ways this can go:
>
> 1) You use phylib. When the PHY reports link, the adjust_link callback
> in the MAC is called. The phydev structure contains information about
> how you should configure the SERDES, SGMII, 2500Base-X, 5000Base-X. It
> works, but it is not so nice.
>
> 2) phylink gives you a much nicer API to do the same. Again, the PHY
> reports the link is up. phylink will then tell the MAC how to
> configure its end of the SERDES. The problem with phylink is that it
> expects a DT building. You don't have that, since this is a USB
> device. But you also don't need a lot of the features of phylink like
> SFPs, the i2c bus for the SFPs, GPIOs etc. So it should not be to hard
> to make this work without device tree.
>
> By using core linux code, we avoid bugs in firmware which nobody can
> fix. The Linux core code should be well tested and supported, but
> phylink is rather new, so might still have some corner cases.
>
> I also cannot imaging parts of the PHY driver will not be re-usable
> for other Aquantia PHYs. I have a board with an AQCS109 under my desk
> waiting for me to do something with it. I really would like a better
> PHY driver for it than the kernel currently has. Hopefully there is
> some code reuse possibilities here.
Even if the firmware is helping, there is still value in using PHYLINK
to report things to Linux as Andrew is saying in that you get a lot of
things for free: auto-negotiation results, link_ksetttings_{get,set} etc.
--
Florian
