R32.3.1 tx2-4g How to run flash.sh on read only file system

[terrysu50z]

Releasing software to manufacturing, flash.sh is ending up running on a read only file system, and of course it is failing.

basically I want to run sudo ./flash.sh -r jetson-tx2-4GB mmcblk0p1 and have it only send files/data to the tx2, not rebuild stuff, I can get it to not rebuild system.img, but it looks like the dtb is being modified.

Terry

[linuxdev]

I’m not positive, but I think what you are seeing is probably content being copied from the reference copy to the “working” copy in “Linux_for_Tegra/bootloader/” and “Linux_for_Tegra/kernel/”. I don’t know if that is 100% the case, but probably it is at least part of it. Since the content is not actually used to populate the “rootfs/” when “reusing” the image, it probably means that most of that content is not actually used in the sense of not actually modifying the “rootfs/”. Could you do either of these?

• Experiment with moving the “bootloader/” and “kernel/” (as a whole, preserving permissions and such without just copying) to a new read-write directory, and then creating a symbolic link to this from the “Linux_for_Tegra/” location so that flash software uses the alternative locations?
• Or you could mount an alternate partition to the “bootloader/” and “kernel/” locations which is read-write (and this in turn, if you wish, could be a loopback mounted file pretending to be a partition).

Both of those would of course need to be the “reuse system.img” method to avoid recreating the image file from “rootfs/” (which in turn is not what you want since this too will modify “rootfs/boot/”).

[terrysu50z]

I think flash.sh rebuilds the dtb from scratch every time before down loading it to the tx2,

What is needed is a “build static” and “load static” that way I could build static, tar and give it to manufactuing, and they would do load static. based on all static files, Nvidia has the -r that uses a static copy of the system.img, Nvidia need to complete the task and do the build static.

[linuxdev]

I have never tried this, but it seems there is a possibility of editing “flash.sh” such that if “-r” is used, then it would not rebuild the device tree (which is probably on one of those subdirectories I mention). It would be logical for NVIDIA to add that modification, but meanwhile, you could make that modification as well and it would be available sooner (I also don’t know which specific release you are using, and if NVIDIA adds that modification for “-r”, then it would likely only be available on a new release).