Bug#587999: further info

[Michel Dänzer]

On Son, 2010-07-11 at 13:43 +0200, Ulrich Eckhardt wrote: 
> On Sunday 11 July 2010 10:05:08 Brice Goglin wrote:
> > What about the original problem (pixel columns switched when using XV,
> > with 16 or 24 bits color depth), is it gone now?
> 
> No, not really, but it doesn't affect me now.
> 
> If I understand correctly, what happened was that KMS broke DRI for X,

radeonfb prevents KMS from working, and the radeon kernel module
currently reacts to this by not initializing at all (instead of
initializing without KMS).

> so it switched back to non-accelerated rendering.
> Non-accelerated rendering gets the pixel columns wrong.

Replace 'Non-accelerated' with 'Non-DRI'.

> In addition, it seems that DRI with 24BPP requires more memory now
> that I have (can someone confirm/explain that?), which made 
> diagnosis more complicated.

This is because by default, when KMS is disabled, the X radeon driver
sets the maximum desktop size such that it's possible to use rotation.
If you don't need that, you can use the Virtual directive to reduce the
amount of VRAM needed for the DRI.

(These issues should apply less if at all with KMS enabled - have you
tried disabling radeonfb instead?)


> After turning off KMS and upgrading, I was to switch back to 16BPP, which 
> works accelerated now, so the bug doesn't affect me. However, this is rather a 
> workaround than a fix.

Indeed.


-- 
Earthling Michel Dänzer           |                http://www.vmware.com
Libre software enthusiast         |          Debian, X and DRI developer