Re: [PATCH] add virtual PHY for PHY-less devices

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

(please do not top-post)

On 9/17/2020 5:17 PM, Sergej Bauer wrote:
> Hi Andrew
> 
>    To tell the truth, I thought that fixed_phy is only for devices with a Device
> Trees and I never met DTS on x86 machines...

fixed_phy existed long before Device Tree was popular and you can 
register a fixed PHY device from within your Ethernet MAC driver, see 
drivers/net/dsa/dsa_loop.c for an example.

> 
> So it looks like there realy no any significant advantage _except_ of
> ability to use ethtool and ioctl to set speed and rx-all/fcs flags without
> removing module. That was most wanted request from HW designers as they are
> wanted to change registers of virtual PHY on-the-fly with ethtool either custom
> tool (using SIOCSMIIREG ioctl) for controling PHY registers.

NETIF_F_RXALL and NETIF_F_RXFCS are feature flags that are applicable to 
the Ethernet MAC, so I am not really sure where the virtual PHY, or the 
fixed PHY plays a role in those?

The key thing with the fixed PHY is that it is an emulation device, 
therefore from the Ethernet MAC perspective you can tell 
(phy_is_pseudo_fixed_link) whether this is a real PHY device on a MDIO 
bus or the emulated fixed PHY but there is really no need for you to 
make that difference.

> 
> p.s. And that's my bad, the original driver was developed year ago (for
> linux-5.2.15),
> but I had no time before this moment.
> 
> 
> p.p.s. sorry for long time to answer but it's far behind the midnight in my
> region.
> 

-- 
Florian