Jetson Orin AGX PCIe Endpoint mode ping issues

Robotics & Edge Computing Jetson Systems Jetson AGX Orin
1

Hello,

I make some experiments with two Orin AGX connected via PCIe and have a few issues. First of all I use the latest Jetpack 6.2 on both sides and configure Ethernet interface via PCIe. Everything works fine, iperf3 gives me 7.05 Gbits/sec bitrate when EP is a server and about 4 Gbits/sec when RP is a server. However, I am completely not satisfying with ping results:

rtt min/avg/max/mdev = 1.789/3.090/3.838/0.923 ms

Here is a ping output from RP to EP and I get almost the same results when I run ping from EP to host.

Is it correct? Are there any limitations why I get a PCIe ping much higher than even a direct ping via Ethernet cable?

The second issue I have is when I turn off or reboot the EP my RP kernel freezes and crashes with this dmesg log:

[  999.550259] pcieport 0005:00:00.0: AER: Uncorrected (Fatal) error received: 0005:00:00.0
[  999.550280] pcieport 0005:00:00.0: PCIe Bus Error: severity=Uncorrected (Fatal), type=Transaction Layer, (Receiver ID)
[  999.550283] pcieport 0005:00:00.0:   device [10de:229a] error status/mask=00000020/00400000
[  999.550285] pcieport 0005:00:00.0:    [ 5] SDES                   (First)
[  999.550295] tvnet 0005:01:00.0: AER: can't recover (no error_detected callback)

Are there any ways to fix this?

5

Sorry for the late response.
Is this still an issue to support? Any result can be shared?

6

Hello,
Nope, the issue has not been resolved and needs to get support

7

  1. The ping result is expected. This has been discussed with other users before. The overhead is from the TCP/IP stack itself which may not be enhanced from PCIe side.

  2. Is that easily reproduced of that AER error? We need to use NV devkit to reproduce and check if it is easily reproduced.

8
  1. Thanks!
  2. It happens every try, at least at my setup (AGX Orin Devkit + AGX Orin Devkit)