Xavier NX Jetpack 5.1.2 makes ethernet pause storm on devkit after soft reboot

We are using Jetpack 5.1.2 on Jetson Xavier NX production module.
The system is running well on ethernet after power on .
When we restart the Jetson Xavier NX using “sudo reboot” it comes up on the serial console fully normally. But it can’t be accessed by ethernet.
This is the case for the NVidia developer kit baseboard and our custom baseboard.
Wireshark shows storms of pause packets as described in AGX Xavier Devkit causes pause frames storm .
This behaviour can be 100% reproduced on every soft reboot.
It might be the case that it also happens after longer times after a power-on-reboot ( maybe 6…8 hrs or so )

Do you have a dmesg to share here?

Xavier AGX ethernet PHY are different from XNX, so the situation might not be same.

Dmesg_EJB_OK_PwrOn_20231214.txt (334.1 KB)
Dmesg_EJB_PAUSEFramesAfterReboot_20231214.txt (246.7 KB)
Hello Wayne,
I uploaded the output from the serial console for a normal start after power on and one for a software reboots ending in the storm of pause frames .
Furthermore I added a screenshot of Wireshark showing the pause fromes sent by Jetson Xavier NX running JP 5.1.2.
Wireshark starts recording pause frames at about timestamp [12.000] to [15.000]. Maybe around first “[ 13.127455] using random self ethernet address”.

FW_EJB_dmesg_after_pwron_20231218.txt (70.4 KB)
FW_EJB_dmesg_after_reboot_20231218.txt (70.9 KB)
Serial_EJB_PAUSEFramesAfterReboot_20231218.txt (83.8 KB)
I made a sudo apt-get update to supress some messages from my last post an made a new test with sudo reboot and power-on. The sudo reboot - case ended again in a pause-storm on ethernet The attached files are from serial output and dmesg for both cases. There are some little differences after “nvethernet 2490000.ethernet eth0: Link is Up” in the dmesg-logs at around timestamps 14s / 17s.
Brgds A.

Serial_EJB_PAUSEFramesAfterPwrOn_20231218 .txt (83.6 KB)

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

I found the reason for the PAUSE frame storm right now:

I mistakenly used tegra194-p3668-common.dtsi from Jetpack 5.0.2 as a template for my changes of the device tree ( i.e. to enable gpio-based chip-selects for spi@3210000 to control two IMX174 sensors and two Crosslink-FPGAs and an SCA3300 inclination sensor).
But I used it for the kernel compilation of Jetpack 5.1.2. . I suppose , that the change from 5.0.2. to 5.1.2 around “ethernet@2490000” caused the problem.
So: Problem solved.
Reason: Version conflict in device tree.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.