[Intel-gfx] drm/i915 X crashes on 835GM (IBM R31)

[Chris Wilson]

On Tue, Oct 22, 2013 at 03:24:18PM +0200, Thomas Richter wrote:
> Am 22.10.2013 14:03, schrieb Chris Wilson:
> >Ah, /var/log/Xorg.0.log has greater verbosity than stdout/stderr. I am
> >just curious where we reject the -intel driver and hence fail to init
> >the screen.
> 
> /var/logXorg.0.log is zero bytes long at this point. I wouldn't call
> this "greater". (-;
> Let it be as it is, I wouldn't consider this an acceptable reaction.
> >>later when X is running, I do get a working X, but no picture on the
> >>external monitor. The xrandr output is also attached.
> >> From the xrandr, I can see the problem. LVDS1 can be on either CRTC, but
> >VGA can only be on CRTC-0. At initialisation, LVDS1 is on CRTC-0 and so
> >prevents the VGA from finding a CRTC.
> >
> >If you do:
> >xrandr --output LVDS1 --off
> >xrandr --output VGA1 --preferred
> >xrandr --output LVDS1 --preferred
> >
> >it should work, or more compactly:
> >
> >xrandr \
> >   --output VGA1 --preferred --crtc 0 \
> >   --output LVDS1 --preferred --crtc 1 --primary --left-of VGA1
> >
> >(adjust primary and position to taste)
> 
> Sorry, not really. After a "xrandr --output LVDS1 --panning
> 2048x1536" on the internal screen, I did a
> 
> $ xrandr --output LVDS1 --off && xrandr --output VGA1 --preferred &&
> xrandr --output LVDS1 --preferred
> 
> and the result was only:
> 
> X Error of failed request:  BadMatch (invalid parameter attributes)
>   Major opcode of failed request:  149 (RANDR)
>   Minor opcode of failed request:  29 (RRSetPanning)
>   Serial number of failed request:  26
>   Current serial number in output stream:  26

If I understand the symptoms right, that should be fixed if you use
SNA instead of UXA. Basically it thinks the intermediate stage exceeds
the hardware limits.

> So apparently, xrandr had problems carrying the panning parameters
> over. Then, following that, I did a
> 
> $ xrandr --output LVDS1 --preferred
> $ xrandr --output VGA1 --preferred
> 
> The result is a working LVDS display, but a nonworking external
> display (non-stable, as if the sync is incorrect, but a signal). I'm
> attaching the output of xrandr --verbose in this specific state.
> 
> Then I did a
> 
> $ xrandr --output VGA1 --off
> $ xrandr --output VGA1 --auto
> 
> and then I got a stable and working external display.
> 
> Anyhow, something's screwed in the display setup. a) xrandr should
> carry over the panning parameters just correctly instead of failing
> and, b)
> it should better get the sync timing right the first time, and not
> the second. (Or actually third).

That's a kernel bug, somewhere.
 
> The "flicker on panning" remains here, but only on the external but
> not on the internal.
> 
> >
> >Flickering is another problem altogether (who votes for DSPARB?).
> >-Chris
> >
> Likely so... But there is at least still a problem with getting the
> timing right on the external display.

But at least we are reducing it down to known bugs!
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre