Hello,
We have custom hardware with a DM648 DSP and a Marvell 88E111 Ethernet phy but have been unsuccessful in getting Ethernet working. We only have one phy rather than the EVM’s two, but the same schematic for the link as the Lrytech EVM board so had expected it to work. Our testing has involved using TI’s benchmark nonCopyTCP testee project and jperf to provide a quality measure.
The EVM board always replies to a standard ping request and jperf results are consistently good.
With the first revision of our HW there were sections of the layout where the SGMII tracks (operating at 1.25GBaud) for Tx and Rx were on top of each other. With this design we always achieve link but the jperf results are erratic and vary from board to board, much slower throughput than the EVM also.
With our second revision of the board the SGMII lines are not on top of each other and don’t cross. The schematic for the Ethernet section are identical for our rev1 and rev2. The rev2 board will establish an Ethernet link but will not reply to a ping.
I have added extra debug to the code confirmed that we have control of the Phy via the MDI lines (I can set the LEDs on the RJ45 connector on and off from register writes and read back of the writes are correct). I have read all registers from the Phy and run identical code on all 3 boards and the register settings are identical.
I have also added debug to log when packets are received from the Phy and transmitted. On both the EVM and our rev1 board I can see packets going in and out, on the rev2 board no packets are received. Again identical software run on the 3 boards.
Looking at the SGMII tracks on the PCB layout we’ve noticed that the positive and negative tracks of each differential pair aren’t identical lengths on our boards:
Lyrtech: 0.315mm length difference between RX P and N; 0.257mm length difference between TX P and N
Rev 01: 0.207mm length difference between RX P and N; 1.005mm length difference between TX P and N
Rev 02: 1.563m length difference between RX P and N; 2.732mm length difference between TX P and N
Could this be the problem? Any suggestions on how to investigate the reliability of the SGMII lines would be much appreciated.
Many thanks in advance, Simon