Unsynced TMDS with xf86-video-intel on 945GM

[René Rebe]

On Wednesday 16 May 2007 22:57:57 you wrote:
> On 5/16/07, René Rebe <rene at exactcode.de> wrote:
> > On Wednesday 16 May 2007 21:36:22 Keith Packard wrote:
> > > On Mon, 2007-05-14 at 09:18 +0200, René Rebe wrote:
> > >
> > > > However often (like about 80% of the X starts) I get flickering and rolling thru
> > > > pixel trash on the external display. It "looks" like there is some sync (or so)
> > > > missing or imprecise, as it's the actual desktop pixel content flushing and
> > > > shuffling by.
> > >
> > > > This can be fixed by re-enabling the LVDS and disabling it again:
> > >
> > > Sounds like the mode programming has some stability issues; the
> > > registers in the chip must be programmed in the right sequence and with
> > > the right delays between various steps in the process. Failing to wait
> > > long enough for a PLL to lock can result in 'bad' behaviour like this.
> > >
> > > About the only way to debug this is to add delays at various points in
> > > the mode setting operation, most likely the CRTC timing programming,
> > > although it's possible that the SDVO programming is broken instead.
> > >
> > > One thing to always try is the latest development version of the X
> > > server and intel driver; there are a few minor fixes which may be
> > > related. Plus, patches you come up with won't need to be ported forward.
> >
> > Do you have an explanation why the distortion on the TMDS-1
> > is mostly disappearing when I rotate the 3D desktop cube?
> >
> > I would rather expect the "mis-programmed" TMDS output
> > to have a static noise than changing with 3D engine load, ... ?
> >
> 
> Probably memory controller contention.  Some cheap radeon cards with
> slow memory have the same problem when 3D is active.  The memory
> controller can't service both clients (crtc and 3D) fast enough.  You
> may try reducing the timing on your panel or, if you can, bump the
> crtc's priority in the memory controller.  This will come at the
> expense of 3D performance.

Yes I know about such memory bandwith configuration and
saturation scenarios, especially with shared memory.

Though in my case it's the other way round: With the rotated
cube the image is stable while the plain static 2D desktop the
image is rolling and flickering over the screen.

-- 
  René Rebe - ExactCODE GmbH - Europe, Germany, Berlin
  http://exactcode.de | http://t2-project.org | http://rene.rebe.name