Well, it is just patchwork, so far mainly stealing bits from other places, nothing too much creative. I have not written anything big by myself.

What's holding me back:

- little or no documentation for the hardware
TI and OMAP architecture is not very open and so is other hardware (wlan chip), we'd like to have documentation for jazelle, 3d chip, IVA, or source for DSP kernel running on tablets so we can add stuff and rebuild it

- fear of breaking the hardware
with no documentation some changes may damage the device, also we still have no info about cold flashing the device, I would probably do more work in hacking the bootloader if I had a way to recover from breaking bootloader or config partitition. Even getting the pinout (for newer devices, 770 is known) and building the right cable takes time.

- chronic lack of time
I'm picking lowest hanging fruit first, there is simply no time for bigger stuff, any small roadblock and priorities change. Over the years there were a lot of comments from @nokia.com people on the mailing list about how everything is easy and right under our nose etc. Not true when seen from outside. Also why to bother reverse-engineering stuff? Yes it is fun for the first or second time but my happy reverse-engineering period was at the time when I had C64 or Amiga, not today. This also relates to

- hate to do duplicated or useless work
Who knows what Nokia is cooking in the labs? Why to bother with reinventing the wheel? Why to bother when next release may break it? The only device worth of attention now because of this is 770 :-)

As for pushing upstream (where upstream is Nokia, not kernel.org), yes this is problem too. Why to bother when Nokia will not include it in stock firmware and most users will not be able to use it. We were thinking/talking in maemo-developers about adding some nice features (mainly pixel doubling, rotation) of the epson video chip to the kernel and optimizing SDL for it, the answer was something along the lines - no we don't want it because it is HW specific hack and this HW feature may not be available on future devices (which is probably true).

Also as Nokia is doing relatively good job with kernel one reason why there is not much work done is simply because there is no need, it already works :-)