Details:

The CPSW SGMII module outputs up to 5 bytes of 0x50 preamble data when in 10/100 mode and there is an odd number of clocks between packets. All bytes should be 0x55. In 1000Mbps mode, which does not have the issue, there are seven 0x55's in the preamble previous to the SFD. In 100Mbps mode there are 70 bytes in the preamble before the SFD (because the data is replicated 10 times from 1000Mbps mode). The first five bytes of the seventy can be 0x50 when the issue occurs. This issue has been undetected until now due to testing only with the PHYs that allow the preamble to be eroded and don't care about the actual data in the first number of bytes However, this issue was recently detected with a Marvel PHY (88Q1111 or similar) that looks at the preamble data and makes packet keep/discard decisions based on preamble data of 0x50.

Workaround(s):

The workaround options are:

1. Use 1000M mode which does not have the issue.

OR

2. Use a TI PHY (DP83869 or similar) or any other PHY which can erode/ignore the preamble data in 10/100/1000M mode.