Strange registration messages

[Sebastian Roth]

@george1421 said:

Putting a dumb mini-switch in between the building switch and the target device broke (blocked) …

Actually it was the other way round!! We couldn’t get the NIC up (LED stayed off) whatever we tried. Then I suggested to put a mini switch in between. And that made it work right away! What is really strange about the whole thing is that iPXE (as well as PXE ROM) can actually bring the NIC up, send packets and receive an IP… Only the kernel seams to have an issue when bringing up the NIC. But as soon as you use dump network equipment it seams to work. I still wonder if this is related to spanning tree or something else?!

[george1421]

@Sebastian-Roth No I still think I’m thinking right. On the house or building switch you could not wake up the port. If the nic and the building switch support 802.3az then they will negotiate a low power mode. But if you plug in a switch that doesn’t support 802.3az the discussion will end and the port will remain in high power mode the whole time.

[Wayne Workman]

@aruhuno Could also look through the firmware settings on the target host and disable low power sleep modes.

[Sebastian Roth]

@george1421 I do understand what you are saying and it sounds logical to one end. But why does PXE boot work ??? Well PXE ROM ok (BIOS magic) - but iPXE ???

Edit: Hmmm, maybe I am looking at it the wrong way round. Possibly the kernel r8169 driver actually is 802.3az/EEE capable and puts the NIC in power save mode. Sorry if that’s what you meant all the time. Didn’t get to my mind until now. There is EEE mentioned in comments a couple of times in that driver - but not very obvious looking to me. Maybe it’s in the firmware files (closed source).

[george1421]

@Sebastian-Roth The only thing I can think is that iPXE is not that smart to support 802.3az protocol but the 4.x linux kernel drivers are. That may explain while the older linux kernels worked on this box (because they did not support this feature).

[aruhuno]

@Wayne-Workman
This is disabled since Tom to me suggested in a previous post.

@Sebastian-Roth
Unfortunately, as I said yesterday , Spanning Tree is active on all switch ports:

[Sebastian Roth]

@aruhuno See my chat messages! For further references and other users: Check out your switch manual! In case you have the same DELL switch take a look here (search for EEE - e.g. page 499). Thanks Tom for pointing this out and George for bringing it up again so I understood as well…

[Sebastian Roth]

Playing with the switch settings it seams to be the auto-negotiation that is causing the problem in this case. Disabling EEE on the client port did not help unfortunately. Then I remembered reading about auto-neg issues here, here and here. Some have less speed and some have no connection at all. Those NICs are not my friend I suppose!

@aruhuno Please check your cabling again! This might help but I am not sure…

[george1421]

So turning off green power has no impact. What happens if you disable auto speed negotiations at the switch? There is no doubt that these network adapters are trouble.

Did we ever have the OP boot using a commercial live distribution of linux to see if we have the same results as with FOS?

[Tom Elliott]

@george1421 This leads me to believe this is NOT an 802.3az issue, but rather something else (STP that is improperly setup for Portfast? Switch module is going bad?

In my case of the 802.3az as the issue I DID see a delay in booting to PXE/TFTP.

[aruhuno]

Following a discussion with @Sebastian-Roth, we finally found it!
The problem is the speed negotiation with the switch . I’ll update the firmware of the switches and make feedback.

[Wayne Workman]

@aruhuno said:

Following a discussion with @Sebastian-Roth, we finally found it!
The problem is the speed negotiation with the switch . I’ll update the firmware of the switches and make feedback.

#wiki