This NVIDIA AGX Xavier board is pretty finnicky! The toolchain for the SDK appears to be weak when using nvme as the main “external” storage. Let me expalin a bit further …
Through the GUI sdkmanager there appears to be a call to l4t_initrd_flash.sh which continues to specify the mmc default, instead of the drop-down specified nvme external device.
Once rebooted to the flashing step here it will hang using sdk manager 1.9.1.10844 on Ubuntu 20.04.
It is possible to ssh to the Xavier over usb with this :
ssh root@fc00:1:1:0::2
password is “root”.
You will have to accept the ssh key for the first time you login to the machine – this may also be holding up the ssh command to flash to the device (the terminal for sdkmanager is unclear and does not report the ssh key-accept message)
Once inside, you can get the device online using ifconfig, assuming you are connected over ethernet and the local LAN has DHCP :
ifconfig eth0 up
dhclient eth0
ifconfig -a
Then you’re online, with the IP shown.
But, it is NOT clear how to progress and flash the generated image to the nvme drive.
Now that the Xavier is here - what command must I run to properly flash it ?
I’ve tried this:
sudo ./tools/kernel_flash/l4t_initrd_flash.sh --external-device nvme0n1p1 -c ./tools/kernel_flash/flash_l4t_external.xml --erase-all --network enp0s3:10.0.2.15/24:192.168.2.57 --showlogs jetson-xavier nvme0n1p1
local LAN is behind a NAT which is the reason for the 10.x address, and 192.168.2.x is what the NAT connects to. (should just work!)
After building the image, I get a :
“Finish generating flash package.
No devices to flash”
How can it not find the device to flash ?