I got a Mikrotik RB2011UiAS-RM from a friend, which I use to port OpenWrt to.
One of the 100M ports (using AR9344 built-in-switch) on the device is misbehaving, establishing a phantom link on 10 megabits. I tried measuring all the coupling elements (resistors, capacitors) at the port, but they all seem fine.
When I touch the port leads on the bottom of the PCB in the right place, or the coupling transformers, this sometimes disappears, and if a device is connected to the port at the time, it allows it to work - of course until next link drop. When the phantom link is up, the port is working as a loopback, needless to say.
Any suggestions on what might be the cause? I suspect some kind of crosstalk between TX and RX, but can't pinpoint where it happens.
What I haven't touched yet is lifting AR9344 itself or the magnetics for 100M switch - the part handling it is a single through-hole unit handling all 5 ports. Adding load resistors to RX line of various values wasn't able to consistently help either - my body was probably working as an antenna and it's the noise that was helping.
How is your soldering? Reads like it is nothing out of your skillset.
You can either trace every line on the port for continuity to the second circuit device looking for a bad solder. Or test the closest resistors, diodes and capacitors individually on each trace out of the port.
Hopefully it is a weak solder. But I suspect you have already considered doing this and just want to hear it is not a bad idea.
Or it might be something like water damage causing a short. Which explains 10M instead of the expected 100M.
No signs of water damage there. I've reflowed every single joint belonging to that port, the passives and even tried reflowing AR9344 with no avail.
The issue is not a downshift to 10Mbps, but a ghost link the port establishes with itself when nothing is plugged in - preventing another devices from connecting, and needing isolation of that port to avoid loops, too.
I'd like to diagnose that to learn the root cause, so I know how to deal with that the next time. I think the right thing to do now is to take every passive component off and measure it. Won't be pleasing, because the components are 0402 size.
Frankly, a schematic for AR9344 reference design would be helpful too, so I can take a look at components wired to the built-in-switch ports. They are likely to be copied verbatim on RB2011.