Forgot to mention some possibly important details.  It's Intel G41 video onboard a Dell Optiplex 380.  The monitor is a Samsung 2494SW, connected via DVI.  

I have kernels built from the commits right before & after the breakage, let me know if you need info from either.

A camera would be good so that I know what type of artifact is being generate. Can you also grab an intel_reg_dumper with working/blurry kernel just to confirm that the registers are indeed the same and it is "merely" a timing issue whilst bringing the panel up.

Created attachment 40616 [details]
photo of blurry vs crisp

This is the best photo I can get for now.  The top of the pic is the blurry/broken one, the bottom is working/crisp.  You can see quite a difference on the words "login" and "ben" especially.

I don't see intel_reg_dumper anywhere after building intel-gpu-tools, is it still named the same?  I do have intel_gpu_dump, and its output is identical on either kernel (confirmed via diff).

Horizontal sub-pixel interpolation. Looks like the panel-fitter is incorrectly programmed.

intel_reg_dumper is not available in the tarball but can be found in the git repo: http://cgit.freedesktop.org/xorg/app/intel-gpu-tools

Thanks.

Created attachment 40618 [details]
intel_reg_dumper output from bad kernel

Created attachment 40619 [details]
intel_reg_dumper output from good kernel

$ diff intel_reg_dumper-bad.txt intel_reg_dumper-good.txt 
44,45c44,45
<         PFIT_CONTROL: 0x80000000
<      PFIT_PGM_RATIOS: 0x10001000
---
>         PFIT_CONTROL: 0x00000000
>      PFIT_PGM_RATIOS: 0x0d551000

Ben, can you also add the output of xrandr? For gen4 hw we should not be touching pgm_ratios at all (so I presume the 1:1 scale-factor there is a residual value from the BIOS). The difference appears that we now always enable auto-scaling for gen4 with the aspect-preserving fullscreen mode.

Created attachment 40637 [details] [review]
Disable pfitter for 1:1 scaling

$ xrandr
Screen 0: minimum 320 x 200, current 1920 x 1080, maximum 8192 x 8192
VGA1 disconnected (normal left inverted right x axis y axis)
DVI1 connected 1920x1080+0+0 (normal left inverted right x axis y axis) 510mm x 287mm
   1920x1080      60.0*+
   1280x1024      75.0     60.0  
   1152x864       75.0  
   1024x768       75.1     60.0  
   800x600        75.0     60.3  
   640x480        75.0     60.0  
   720x400        70.1

Sorry, actually wanted xrandr --verbose, so I can see which scaling mode is active for the panel.

Created attachment 40640 [details] [review]
Disable pfitter for 1:1 scaling

/me slaps himself.

I missed the obvious, there is no LVDS on these systems - the panel fitter should be disabled in these cases.

Created attachment 40642 [details]
xrandr output

xrandr output is the same on bad vs good kernels (except for 2 timestamp lines)

Created attachment 40643 [details] [review]
Only use pfit with lvds

I'm not going to pretend to know what's going on in that last patch(0001-drm-i915-Clear-pfit-registers-when-not-used-by-any-o.patch), but it does seem to fix the problem for me, applied against vanilla 2.6.37-rc3

Ben, thanks a lot for the report and testing.

I've queued the fix for upstream:

commit c5d1b51d3559664920136b45f4d2366ed9a9e8be
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Mon Nov 29 18:00:23 2010 +0000

    drm/i915: Clear pfit registers when not used by any outputs
    
    ... otherwise the panel-fitter may be left enabled with random settings
    and cause unintended filtering (i.e. blurring of native modes on external
    panels).
    
    Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=31942
    Reported-and-tested-by: Ben Kohler <bkohler@gmail.com>
    Tested-by: Ciprian Docan <docan@eden.rutgers.edu>
    Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

and will push as soon as kernel.org is up again.