Hi,
We are in the process of developing a complex system.
This system has two embedded boards: Board A and Board B.
Board A is connected directly to Board B through 1000BASE-T ethernet (using a CAT 6A twisted pair cable).
Connection speed is 1 Gigabit.
Board B is always powered on.
Board A is initially powered off, and when we power it on, it needs to establish ethernet connection in less than 2 seconds.
As far as I understand, auto-negotiation is mandatory for gigabit ethernet, and this process may take more than 2 seconds.
I performed a few experiments using two Xilinx evaluation boards (ZCU102).
These boards contain a TI DP83867IRPAP Ethernet PHY.
To reduce the time to establish Ethernet link, I planned to enable the (non IEEE standard compliant) Fast Link Detect mode (also called "Fast Auto-Negotiation) of the PHY's at both ends, and set one end to Master and the other to Slave.
I saw that that this configuration works in about 75 percents of Board A power cycles:
In about 75 percents of the power cycles, Ethernet link is established in less than 1.5 seconds from power on (Note that the the Xilinx ZCU102 board deasserts PHY reset a few hundred milliseconds from power on - controlled by software).
However, in about 25 percents of the power cycles I saw that Ethernet link is established only after about 2.7 seconds from power on.
Note that the software that I run on both boards is a minimal bare metal software.
I don't know whether or not the following is relevant: I tried to dump the PHY registers on board B, and noticed in the slow link up cycles that the STS1 register contains the value 0x08FF, while the Link Status field in BMSR drops to 0.
Any help will be much appreciated.