On 27.07.2021 18:32, Marek Behún wrote:

Hi,

On Tue, 27 Jul 2021 17:53:58 +0200
Michael Walle [off-list ref] wrote:

quoted

quoted

If we used the devicename as you are suggesting, then for the two LEDs
the devicename part would be the same:
  ledA -> macA -> ethernet0
  ledB -> phyB -> ethernet0
although they are clearly on different MACs.  

Why is that the case? Why can't both the MAC and the PHY request a 
unique name from the same namespace?

So all the network related devices should request a unique network
relate device ID? Should also wireless PHY devices do this? WWAN modems?
And all these should have the same template for devicename part withing
/sys/class/leds? What should be the template for the devicename, if
wireless PHYs and WWAN modems could also be part of this? It cannot be
"ethernet" anymore.

It seems a better idea to me to just some nice identifier for the LED
controller.

quoted

As Andrew pointed out, the names in
/sys/class/leds don't really matter. Ok, it will still depend on the
probe order which might not be the case if you split it between ethmac
and ethphy.

Yes, the LED name does not matter. But the LED subsystem requires names
in a specific format, this is already decided and documented, we are
not going to be changing this. The only reasonable thing we can do now
is to choose a sane devicename.

quoted

Sorry, if I may ask stupid questions here. I don't want to cause much
trouble, here. I was just wondering why we have to make up two different
(totally unrelated names to the network interface names) instead of just
one (again totally unrelated to the interface name and index).

It seems more logical to me from kernel's point of view.

quoted

But I was actually referring to your "you see the leds in /sys/ of all
the network adapters". That problem still persists, right?

Yes, this still persists. But we really do not want to start
introducing namespaces to the LED subsystem.

Marek

Did we come to any conclusion?

My preliminary r8169 implementation now creates the following LED names:

lrwxrwxrwx 1 root root 0 Jul 26 22:50 r8169-led0-0300 -> ../../devices/pci0000:00/0000:00:1d.0/0000:03:00.0/net/enp3s0/r8169-led0-0300
lrwxrwxrwx 1 root root 0 Jul 26 22:50 r8169-led1-0300 -> ../../devices/pci0000:00/0000:00:1d.0/0000:03:00.0/net/enp3s0/r8169-led1-0300
lrwxrwxrwx 1 root root 0 Jul 26 22:50 r8169-led2-0300 -> ../../devices/pci0000:00/0000:00:1d.0/0000:03:00.0/net/enp3s0/r8169-led2-0300

I understood that LEDs should at least be renamed to r8169-0300::link-0
to link-2 Is this correct? Or do we have to wait with any network LED support
for a name discussion outcome?

For the different LED modes I defined private hw triggers (using trigger_type
to make the triggers usable with r8169 LEDs only). The trigger attribute now
looks like this:

[none] link_10_100 link_1000 link_10_100_1000 link_ACT link_10_100_ACT link_1000_ACT link_10_100_1000_ACT

Nice, or? Issue is just that these trigger names really should be made a
standard for all network LEDs. I don't care about the exact naming, important
is just that trigger names are the same, no matter whether it's about a r8169-
or igc- or whatever network chip controlled LEDs.

And I don't have a good solution for initialization yet. LED mode is whatever
BIOS sets, but initial trigger value is "none". I would have to read the
initial LED control register values, iterate over the triggers to find the
matching one, and call led_trigger_set() to properly set this trigger as
current trigger. Most likely this would need some LED core extensions:
- enable iterating over all triggers with a particular trigger_type
- enable triggers to have private data
Quite some hassle for a small functionality.