12-11-2018 08:10 AM - edited 12-11-2018 09:42 AM
Target FPGA : Zynq7z020
Vivado 2016.2, Petalinux 2016.2
Using AXI Ethernet, AXI DMA and those are connected to ACP port of PS. This ACP port is also shared with our custom DMA module (This DMA mostly read from RAM). Jumbo frame is set but the problem still happens in standard frame (1500B). Jumbo frame MTU is 9000B. We connected multiple target boards to one host server PC through a 10G-1G ethernet switch. We are using Marvell 88E1512, RGMII interface for phy chipset.
While running the stress test from PC to multiple target boards (2 or 4), randomly the ethernet connection is lost. This happens very rarely.
1st test : connection drops from one target board at about 240GB data transfer
2nd test : connection drops from one target board at about 470GB data transfer
3, 4th test : connection drops from one target board at about 1.2TB data transfer
RJ45 lights are : Orange is ON, Green is always ON, not blinking. Network switch's light for the port of target board is off. All other ports in network switch are alive. If I change the connection of dropped port with a connection of a good target board, it works. This means network switch works fine.
I checked the software. I connected a serial communication to the target board and can run Linux shell commands. I ran a ethtool, ifconfig, and devmem to check the trimac registers. Software seems fine.
When I check the trimac registers, everything was good. There is no errors, no packet drops. I dumped some registers with ethtool and attached it to this post.
One important thing is that if I ran "ifconfig eth0 down" and "ifconfig eth0 up", then the link is up and everything works fine again.
[Tue Dec 11 08:37:11.758 2018] root@samba_fuji:~# xilinx_axienet 41000000.ethernet eth0: Link is Down
[Tue Dec 11 08:37:15.759 2018] xilinx_axienet 41000000.ethernet eth0: Link is Up - 1Gbps/Full - flow control off
[Tue Dec 11 08:37:15.759 2018] IPv6: ADDRCONF(NETDEV_CHANGE): eth0: link becomes ready
Since after I reset the AXI MAC with ifconfig command, the MAC and PHY works again, I can't dismiss the idea that there is something wrong in the lower layer (layer 1 or layer 2).
Please help us on stabilizing the AXI ethernet connection.
01-09-2019 10:35 PM
Is the same issue seen across all the speeds? I mean did you check with other speeds 10/100Mbps and found link stable?