[directfb-dev] Crude savagefb driver for 2.6 kernels

[Mika Pruikkonen <mpruikko@cc.hut.fi>]

Hi,

You can find a crude patch for S3 Savage framebuffers in:

http://www.hut.fi/~mpruikko/savagefb/patch-2.6.7-savagefb

It's based on the 2.4 savagefb driver found in the directfb project's 
cvs tree, with minor resyncing to the X driver. I really have no clue 
about video driver programming, nor actually about kernel hacking
either, but I wanted to be able to do acpi suspend and have the 
framebuffer work after the resume too, so here's the result. :)

I use this on an IBM Thinkpad T23, which has a S3 SuperSavage IXC and 
a 1024x768 lcd screen, and the driver hasn't been tested on anything 
else, so it will very probably not work without tweaking on most 
savages.

Features: screen corruption during mode switching, interesting text 
scroll effects, refreshingly unusual console color scheme at 16bpp, 
hard-coded boot resolution 1024x768@8bpp, warranty void if you're 
trying to use other resolutions than 1024x768. And I wouldn't try
compiling it as a module.

I also noticed that the driver was noticeably slower than vesafb (e.g. 
using mplayer with fbdev/fbdev2, text scrolling, etc). Following patch 
(incremental) hard-codes some timing registers to values which were in 
use when using vesafb (with vga=792):

http://www.hut.fi/~mpruikko/savagefb/patch-2.6.7-savagefb-hardcoded-timing

With this patch the performance seems comparable to vesafb. I have 
no idea how to fix this in general, though. I converted the timing 
calculation functions to use non-floating-point arithmetic, but I think 
they should produce the same results as in the directfb driver. Also, I 
remember reading that someone complained about worse-than-vesafb 
performance in the directfb driver too. Any ideas?

I hope someone can find this useful and/or maybe even get it to work, 
and I would welcome patches for the known "features" (or unknown ones), 
or much-needed cleanup.. best of all would of course be, if someone
who actually knew what he was doing would produce a proper driver to
be included in the kernel. :)

Regards,

  mp