DP83869EVM: RGMII interface to MDI interface conversion 1Gbps bandwidth not working

Part Number: DP83869EVM
Other Parts Discussed in Thread: DP83869

Wiring connection Explanation:

 

image.png

 

image.png

 

 

Objective: RGMII interface to MDI interface conversion 

We are trying to validate the RGMII interface to MDI interface (RJ45 Connection) at 1Gbps bandwidth.

What Worked (10 Mbps)

Bidirectional ping confirmed at 10 Mbps (forced via ethtool -s eif1 speed 10 duplex half autoneg off)
iperf throughput validated at 10 Mbps
Both TX and RX frames are incremented (tx_good_frames and rx_good_frames both non-zero)
PHY correctly detected at MDIO address 3 (phy_id = 0x2000c3d8)
Link comes up (with flapping) confirms wiring direction is correct

What Did NOT Work (100 Mbps / 1000 Mbps)

At 100M: Link comes up but RX is completely dead — tx_good_frames increments, rx_good_frames = 0
At 100M: Link flaps continuously (Up/Down every ~1 second)

 

Questions:

 

  1. Do we need to install DP83869 driver separately...? or its part of uboot kernal driver...?
  2. What is maximum allowable trace length for the RGMII interface, is this connection length valid...?
  3. Able to see link active state for 10/100/1000Mbps, But it is working ping command only for 10Mbps, verified iperf3 throughput for the same. What is the issue..?
  4. Have Probed both RX & TX clock able to see 125Mhz frequencies.
  5. RGMII interface interrupt pin (INT_PWDN) from DP83869 EVM to processor required to connect to processor...?
  6. MDC & MDIO interface works properly and able to configure Read the PHY address.
  7. What is the auto negotiation boot strap followed for 1Gbps bandwidth.
  8. Is there application notes or forums  to resolve this issue.?
  • Hi Seenivasan,

    1. The current DP83869 driver is located on our public Github.

    2. The maximum trace length we recommend is 6 inches. 

    3. This EVM is not intended for throughput testing (full signal chain). The RGMII routes to probe points/headers on the board that are good for measuring signals coming from the PHY. These probe points are not ideal for a high speed signal (such as RGMII) to pass through.

    4. This is good and indicates you are linked up at 1Gbps

    5. INT is not required to connect to a processor. 

    6. This is good

    7. 1Gbps autonegotiation would be set using the OPMODE straps. You should remove any straps on the OPMODE pins to get 1Gbps Copper to RGMII

    8. We do have tips for troubleshooting the RGMII, however this EVM's RGMII connection is not ideal for signal integrity as it uses jumper wires to connect to a second board. This may be ok for 10Mbps, which is lower speed, however for 100M and 1Gbps this connector likely prevents the signal from passing cleanly. For validating the RGMII interface it would be better to use a board with trace connections or a dedicated board to board connector for the RGMII.

    Best,

    Shane