Having a really strange bug bringing up Ethernet on STARSHIPRAIDER. When I force the link speed on my switch to 10M, it links up and appears to be OK (although my MAC doesn't run in 10/100 mode yet so I can't actually test with TCP/IP traffic)
100M and 1G modes flap constantly and won't stay linked up. Suspecting signal integrity or power problem.
Tried swapping cables and switch ports already, no luck there.
Tried swapping cables and switch ports already, no luck there.
Scope captures of the various 1000base-T test patterns look fine, as do eyes on 100baseTX transmit (I can't do CDR on 1000base-T yet). All power rails appear to be the correct voltages. Visually inspected all solder joints and touched up any questionable ones.
All power pins on the PHY (KSZ9031RNX) are decoupled to ground with 0.47 uF 0402 X7R MLCC practically touching the chip. Each rail has a single 4.7 uF as well. Analog rails are filtered with a pi filter consisting of 4.7 uF on either side of a 600R @ 100 MHz ferrite chip.
Here's my characterization setup for waveform measurements, along with Mode 1 and Mode 2 test patterns on pair A. The mode 1 test looks possibly a bit noisy? Doesn't look this bad on the DSO screen and the mode 2 waveform is fine so IDK? 
Here's pair B in forced 100M mode. Looks fine to me, rise time is a little slow maybe? Why would a PHY keep dropping a link this clean?
I've tested with my Cisco 2970G, a Netgear GS108, and the Realtek PHY on my Zybo. Same results on all 3 - constant flapping.
I've tested with my Cisco 2970G, a Netgear GS108, and the Realtek PHY on my Zybo. Same results on all 3 - constant flapping.
The KSZ9031 datasheet says that it can correct for pair inversions, and to ease routing I swapped P/N on pairs B and D. Just in case this was the root of the problem I made a custom patch cable that un-swapped the pairs. No change in observed behavior.
