Thanks for the clarification!

But still possible put the 3d driver in mer , megoo, etc .. ?

Don't see why not, from the technical standpoint. The license on the other hand is quite problematic.

The closed stuff is platform independent and hopefully relatively stable. TI/PowerVR have same code running on other omap2 based devices both with symbian and linux (hopefully).

The tricky part is understand what mbxaccess.ko and omaplcd.ko (main part that needs fixing IMO) is trying to do and adapt it to N810 specific video driver situation which is a bit complex and different from what current omaplcd.ko is expecting. and we have sources for that.

The first thing I would try it to remove everything that touches DISPC (OMAP builtin display controller) registers (vsync flipping code, power management code), IMO our omaplcd.ko implementation should be purely memory based = only giving proper framebuffer memory to rest of powervr code and not touching any hardware directly. It should also not allocate any extra memory if possible (this is what fails now), omapfb already has three allocated planes/buffers, the last one even has space for two 800x480 frames.

Then there should be hook how to schedule display update via current omapfb. Possibly from current omaplcd vsync code (which currently messes with DISPC registers when trying to switch visible buffer with back buffer).

Hopefully that's all for showing MBX output properly.

Then there is mbxaccess which gives access to 3D chip to user space PowerVR libraries. I'm not sure if this one needs to allocate some additional memory too (for textures?) or needs other fixing.

the kernel part is hopefully not a problem, we should import current code in our own place (garage project?) and work on that freely.

As for the rest, the TI download page is quite bad solution, Stskeeps, is there some plan to make it more public? (No TI registration, move it to maemo.org download like firmwares/other binary blobs)

Count me in, if nothing, at least for testing purposes.

I'm not sure, but let us see where this leads us. The important thing is that the drivers are finally here. For licensing issues we need someone of higher rank than me

Still doesn't work on n810

No, it doesn't

And I just happen to be the "expert" because I roamed around at the right moment with the right book, so we clearly have a dream team here!

One of things I've guessed from the current driver is that in previous incarnations, the mbxdaemon was in kernel-space. My guess is that TI split the driver in two pieces: they made the mbxdaemon with the "cannot-ever-possibly-be-GPL" code, and released all the other parts in the mbxaccess kernel module (the previous, all in one driver was called simply `mbx.ko'). I just hope the performance out of this split is acceptable...

Also other than a mailing list I'd also ask putting the kernel module sources in a repo (they're GPLv2 after all).

All my tests have been done on a N810.