Jetson xavier NX booting problem

Hello, after flashing my Jetson Xavier NX, it freezes almost immediately, even before the kernel loads (serial console output below).

Jetson UEFI firmware (version 6.0-37391689 built on 2024-08-28T08:47:11+00:00)
ESC to enter Setup.
F11 to enter Boot Manager Menu.
Enter to continue boot.
…L4TLauncher: Attempting Direct Boot
EFI stub: Booting Linux Kernel…
EFI stub: Using DTB from configuration table
EFI stub: Loaded initrd from LINUX_EFI_INITRD_MEDIA_GUID device path
EFI stub: Exiting boot services and installing virtual address map…

It’s strange that I don’t see any output from the internal bootloaders (on Orin NX/Nano, the primary bootloaders display output before UEFI loads.

BSP - 35.6.0

Device - Xavier Nx 8 GB, 16GB emmc with external nvme disk with custom board.

Command to flash:

“sudo ./tools/kernel_flash/l4t_initrd_flash.sh -S 70GiB --external-device nvme0n1p1 -c tools/kernel_flash/flash_l4t_t194_nvme.xml -p “-c bootloader/t186ref/cfg/flash_t194_uefi_spi_emmc_p3668.xml --no-systemimg” --showlogs --network usb0 jetson-xavier-nx-devkit-emmc external”

The same command works fine for Orin NX (with its own configs). Could this be related to the custom carrier board? The device was previously flashed and working, but it was using the internal eMMC.

Attached is the l4t_initrd_flash.sh flashing log

xavier_flash.log (182.9 KB)

I tried flashing the internal memory using the flash.sh script, but it hangs at the exact same spot

Command: “sudo ./flash.sh jetson-xavier-nx-devkit-emmc mmcblk0p1”

Custom carrier boards usually require a different device tree (firmware). This wouldn’t be the case if the carrier board has an exact copy of the NVIDIA carrier board, and if the module on that carrier board is the dev kit module and not the commercial module sold separately.

That device tree is often used to pass information about any device which is not plug-n-play. Things like the physical address the device resides at, and the driver which can be used. Flashing using the wrong device tree would never trigger any error, but actual boot would fail to find certain devices (while devices which do match where the device tree says to look still work).

A dev kit version of the Xavier NX does not have eMMC, it instead has an SD card slot on the module itself, and not on the carrier board. I would suspect that you have incorrect firmware related to finding and/or using at least those items. Perhaps flash detected the commercial module, but some other part of the boot hardware will more or less just be “missing” (even if the software looks for the hardware, then the hardware won’t be found).

Third party and custom carrier boards will most likely have that third party’s device tree in their own flash software, or else as a patch to the dev kit flash software. In some cases the manufacturer might say to just use NVIDIA’s software since their board is identical in electrical layout.

Hello, the device was working fine before the flash (it might have been using a slightly modified DTS where some GPIOs were reconfigured for SPI). In any case, the kernel should output something to the console at the very beginning, or perhaps the console is misconfigured. I haven’t changed any configs; I just unpacked the BSP and the sample rootfs into Linux_for_Tegra/rootfs, then ran the apply_binaries.sh script. Both scripts l4t_initrd_flash.sh and flash.sh

Both l4t_initrd_flash.sh and flash.sh scripts finished successfully, but I don’t see anything in the console except for the messages mentioned above. Furthermore, as I understand it, flashing via l4t_initrd_flash.sh uses the same kernel and DTS to flash the external drive on the device itself. According to the log, this kernel successfully booted, initialized the USB network, and wrote to the NVMe drive.

I might have misconfigured something for l4t_initrd_flash.sh, although flash.sh only needs the target board config and the partition to work.

Which of these three configs should I use for the Xavier NX, or what changes do I need to make to them?

  • jetson-xavier-nx-devkit.conf
  • jetson-xavier-nx-devkit-qspi.conf
  • jetson-xavier-nx-devkit-emmc.conf

According to the table on this page - https://docs.nvidia.com/jetson/archives/r35.6.0/DeveloperGuide/IN/QuickStart.htmlhttps://docs.nvidia.com/jetson/archives/r35.6.0/DeveloperGuide/IN/QuickStart.html, I am using exactly the one I specified when running both scripts.

Maybe I need to adjust the CMDLINE_ADD in p3668.conf.common to ensure logs are routed correctly? I’m wondering if the lack of output is due to an early panic or if the console settings just need additional parameters?

Have you gone to the third party carrier board’s web site and checked their flashing instructions? I would expect boot to work correctly only if either (A) you’re using the software they say to use (which might be NVIDIA’s or their own), or (B) the parts used for boot do not require any difference in firmware.

Not being able to see something or load a driver might or might not result in a log message. The scripts won’t do any good if the firmware is wrong because they’d install the wrong firmware.

One thing you are looking at is that conf files have a given name chosen because they are tied to specific hardware (and the device tree telling the kernel how to find that hardware). [rearranged for clarity]

Hi, the issue was indeed related to the serial port (it’s either malfunctioning, connected incorrectly, or misconfigured). The system was successfully flashed and is now running using the ‘jetson-xavier-nx-devkit-emmc.conf’ from BSP 35.6.0. Here are the dmesg lines regarding the serial port:

[0.000000] Kernel command line: root=PARTUUID=9740bb83-eb91-4c8d-b413-b07fc3ccce65 rw rootwait rootfstype=ext4 console=ttyTCU0,115200n8 console=tty0 fbcon=map:0 net.ifnames=0 video=efifb:off nospectre_bhb
[ 2.511099] printk: console [ttyTCU0] enabled

The command line is added to the extlinux.conf file by the script during flashing, and I’m using the tegra194-p3668-0001-p3509-0000.dtb from the same BSP version (35.6.0). It’s unclear why the debug serial port is not working.

I’m glad it works now. The command line has a lot of similarities to what a device tree node does, but it is more general. The console token is passed to the driver as parameter content right as the kernel loads, and so it allows debugging right at kernel load.