On Fri, 24 Jul 2020 12:29:01 +0200
Pavel Machek <pavel@ucw.cz> wrote:

> In future, would you expect having software "1000/100/10/nolink"
> triggers I could activate on my scrollock LED (or on GPIO controlled
> LEDs) to indicate network activity?

Look at drivers/net/phy/phy_led_triggers.c, something like that could
be actually implemented there.

Some of the modes are useful, like the "1000/100/10/nolink". But some
of them are pretty weird, and I don't think anyone actually uses it
("1000-10/else", which is on if the device is linked at 1000mbps ar
10mbps, and else off? who would sacrifies a LED for this?).

I actually wanted to talk about the phy_led_triggers.c code. It
registers several trigger for each PHY, with the name in form:
  phy-device-name:mode
where
  phy-device-name is derived from OF
    - sometimes it is in the form
      d0032004.mdio-mii:01
    - but sometimes in the form of whole OF path followed by ":" and
      the PHY address:
      /soc/internal-regs@d0000000/mdio@32004/switch0@10/mdio:08
  mode is "link", "1Gbps", "100Mbps", "10Mbps" and so on"

So I have a GPIO LED, and I can set it to sw trigger so that it is on
when a specific PHY is linked on 1Gbps.

The problem is that on Turris Mox I can connect up to three 8-port
switches, which yields in 25 network PHYs overall. So reading the
trigger file results in 4290 bytes (look at attachment cat_trigger.txt).
I think the phy_led_triggers should have gone this way of having just
one trigger (like netdev has), and specifying phy device via and mode
via another file.

Marek