Douglas,
I will try to look you up! That may help me figure out faster whether or not this is a lost cause!
The Kernel is coded into NOR flash, and the powers that be are VERY hesitant to change it in any way for fear of bricking another unit. That is despite the fact that a previous firmware rev required an update of the kernel in NOR, so we have complete binary images of the up/down grade and the software to burn them in. Of course, that only works if you can still boot to the OS! If we make a minor mistake, we currently have no way to fix it other than send it back to the vendor to reflash it, and management would be very unhappy at the expense.
From what I can tell, the NOR flash compressed kernel image goes most, but not quite all of the way through the boot process. It does install yaffs so that the final boot code can be read from /boot, but I think it loads more by that point than usual. At the very least, it can output an error message on the serial port if the NAND flash fails. But not enough to run even basic recovery functions like cp or tar.
The ttylinux sounds interesting, and the busybox too. I donwloaded a pre-compiled busybox for the ppc, and it was over 1Mbyte, which is bigger than the ramdisk we have to work with. I was hoping that stripping out almost everything but tar, ssh, scp, cp, mv, rm, md5sum and a few others we could get one small enough to load into the ramdisk after normal system boot and then use chroot to have the system use the ramdisk busybox instead of the normal bin and sbin files.
I worst comes to worst, and it sounds like building a cross compiler is already bad enough, we can also try coping the critical system files to the ramdisk and then using chroot. It may take some trial and error to find out all the files that need to be copied over. But then there would be no need to cross compile, the programs already run on the system. I just wasn't sure which approach would be more difficult and/or risky. But if building a cross compiling environment will take forever and it doesn't take too many files copied over to keep the system running, than using the existing code but running from the ramdisk may be the best way to go. Pity, building the cross-compiler sounded interesting. I just can't get approval for a 6 month science project!