Just a very quick update to those keeping an eye on my struggle with TILER, my second attempt at forcing fbdev into tiler memory involved just forcibly widening it to 8192 pixels, thereby making the buffer physically contiguous and making life much easier. This is obviously a waste of memory but I wanted to see something working first and worry about doing it *properly* later
It actually seemed to work initially, resulting in a rotated framebuffer console. Reports are however that running x11 with the fbturbo backend crashed. I haven't investigated yet. Still, this is considerable progress compared to attempt #1.
Note that drm applications are not automagically rotated by this, although it does make drm dumb buffers allocated in tiler memory hence all that's needed is setting the rotation property. I figured that since the number of drm modesetting clients is rather low it might actually be easier to just have a "natural orientation" udev property or something and fix individual clients rather than handling this in kernel. It doesn't help that omap_bo_new(), unlike the dumb buffer equivalent, simply doesn't give the kernel enough information to force allocation into tiled memory.
Matthijs