Cadence GEM (MACB) - Direct RGMII MAC-to-MAC communication without PHY

Hi Team,

We are working on a direct RGMII MAC-to-MAC communication setup between an NXP i.MX8MP MAC and a Cadence GEM (MACB) integrated in an APSoC. The two MACs are connected directly without an external PHY.

We have already performed the following verification to isolate the issue:

• Verified standalone external loopback on the APSoC by connecting TX and RX locally. Both TX and RX packet counters increase correctly.
• Verified standalone external loopback on the i.MX8MP. Communication is successful.
• Verified the RGMII interface using an oscilloscope. TX data and clocks are present, and the electrical interface appears normal.
• Tried different phy-mode settings (rgmii, rgmii-id, rgmii-rxid) without improvement.

Since both MACs work correctly in standalone testing, the remaining suspicion is related to the direct MAC-to-MAC RGMII interface and clock timing.

We would appreciate clarification on the following:

1. Does the Cadence GEM IP officially support direct RGMII MAC-to-MAC communication without an external PHY?
2. Are there any additional GEM register configurations required specifically for MAC-to-MAC operation?
3. Is the RGMII TX/RX clock skew (internal delay) programmable in the Cadence GEM IP?
4. If programmable, is it configured through:
   • GEM registers,
   • the Linux macb driver,
   • Device Tree,
   • or platform-specific hardware outside the GEM IP?
5. Does the Linux macb driver configure any RGMII delay automatically, or is this expected to be handled by the SoC integration?
6. Are there any application notes, reference designs, or known limitations for GEM MAC-to-MAC operation over RGMII?

Any guidance would be greatly appreciated.