+++ Wookey [2010-05-04 11:49 +0100]:
> +++ Charles Manning [2010-05-03 17:02 +1200]:
> 
> > I still have not managed to run built code on balloon
> Did you try the working binaries I uploaded a while back? I have a
> newer set which is definitely good for nor-kernels case. I can upload
> those too, but best to wait till tomorrow when I can test and be sure
> that what I've uploaded really does work.

An update on where we are at with making a balloon release that works.

The modutils problem is present if you build with
too-new-for-buildroot modutils, so don't build on Debian squeeze - use
lenny.

Apart from that NOR-kernel booting is fine. However chrooting into the
rootfs produces a segfault, so the 'setup' process for configuring a
debian/emdebian rootfs does not actually configure anything. I have
worked out how to use qemu-static to cross-configure the rootfs and
the result seems to boot OK, so I suggest we drop the whole 'configure
on the real hardware via USB stick' palaver, as qemu-static is much
quicker and simpler. I haven't yet check if the lenny veriosn of qemu
is new enough to work - it may well not be. And code to do this is not
yet in the repo.

These unconfigured rootfses were previously confounding tests for what
worked and what didn't. Sorting that out has cleared most of the fog.

Getting to the bottom of why chroot has stopped working would be
worthwhile but is a bit tricky. There is no strace in buildroot and if
you use an older Debian environment to test-chroot it in then nothing
goes wrong. Sigh.

NAND-booting with current kernels invaribale fails with:
VFS: Mounted root (yaffs2 filesystem) on device 31:1.
Freeing init memory: 100K
Kernel panic - not syncing: Attempted to kill init!

However if you use the current TCL production kernel then it boots
just fine. So _something_ has changed in the kernel which is causing
this problem. I'm currently binary-chopping my way through likely
revisions to work out what broke it, as it's not obvious. NickB's
'magic connect kernel' of Jan 8th also works, but I don't know what
code/revision that was built from.

r895 gives a backtrace when it dies in the same place:
(However I'm not convinced that was a clean build, so view that 
datapoint with some suspicion)

In the meantime I've updated
http://balloonboard.org/files/balloon3/distro/rel-v0.99/ with a kernel
and rootfs which boot here (but do have mismatched modules).

Going back to revision 843 (clean build) works OK. So, somewhere
between r843 and r1053 root-mounting broke. 

But now it's bedtime. I'm away (at UDS) next week so unless someone
else wants to play this game there won't be any action until friday. 

Wookey
-- 
Principal hats:  iEndian - Balloonboard - Toby Churchill - Emdebian
http://wookware.org/