Hi,
We have a custom design composed of two boards:
- One "main" board with ethernet connector and magnetics,
- a SOM that attaches to main board that has CPU and ethernet phy.
This design was validated with various main boards and different SOMs. Nonetheless, all ethernet phys we used so far were non-Ti.
Recently, we did a respin of one SOM where we replace non-Ti Phy with a DP83867CS (but same SoC/MAC). Software is already validated with previous spin, apart from Phy-specific part. We use u-Boot/Linux.
DP83867-based SOM connection is unreliable. When established, link drops after a few seconds and autoneg happens again, in a loop.
So far, we checked:
- strapping: looks correct (reg 0x6e = 0x0003, reg 0x6f = 0x0030)
- power supplies: they are as recommended (10uF on rail, 1uF+100nF decoupling per pin), probed them with scope, cannot see any glitch,
- termination caps on center taps of MDI (100nF on each center tap), they are correct,
- oscillator: it is in spec (MEMS oscillator on XI, measured as 25M - 8ppm with precision freq counter, scoped waveform looks good).
Reading STS1 (0x000a) as fast as possible (from u-Boot command line), Idle Error Counter is 0xff most of the time.
Register RECR (0x0015) increments regularily (but slow enough for us to see it increment).
What else could we do to troubleshoot this ? Is there any register we could check for detailed diagnosis of link down ?
Thanks
