Bug#641622: Screen corruption regression

[Philipp Klaus Krause <pkk@spth.de>]

Package: linux-image-3.1.0-rc4-amd64
Version: 3.1.0~rc4-1~experimental.1
Severity: Important

There is a screen corruption regression for intel graphics. It sometimes
appears spuriously and has been observed by multiple people. However
there are ways way to reproduce it:

In the free game simutrans, in fullscreen mode (install it from teh
Debian package, start using "simutrans -fullscreen") I see the following
problem:
The lower about 2 cm of the screen don't update. They are black or carry
the image from the game startup screen. They flicker once in a while,
but otherwise the image stays the same.
This makes the game practically unplayable in fullscreen mode.

lspci output:
00:02.0 VGA compatible controller: Intel Corporation Mobile GM965/GL960
Integrated Graphics Controller (primary) (rev 0c)
00:02.1 Display controller: Intel Corporation Mobile GM965/GL960
Integrated Graphics Controller (secondary) (rev 0c)

glxinfo:
OpenGL vendor string: Tungsten Graphics, Inc
OpenGL renderer string: Mesa DRI Intel(R) 965GM
OpenGL version string: 2.1 Mesa 7.11

I have bisected this issue to the following commit:

d74362c9e45689d8d7e3d4bcf6681c4358ef4f2e is the first bad commit
commit d74362c9e45689d8d7e3d4bcf6681c4358ef4f2e
Author: Keith Packard <keithp@keithp.com>
Date:   Thu Jul 28 14:47:14 2011 -0700

    drm/i915: Flush other plane register writes

    Writes to the plane control register are buffered in the chip until a
    write to the DSPADDR (pre-965) or DSPSURF (post-965) register occurs.

    This patch adds flushes in:

    	intel_enable_plane
    	gen6_init_clock_gating
    	ivybridge_init_clock_gating

    Signed-off-by: Keith Packard <keithp@keithp.com>

:040000 040000 68666ff438ff2b9890a72914905b4aadf53e23ef
101e78df34814a3661c87e9e44d10761d0676d7a M	drivers

Switching to the console and back or taking a screen shot is a temporary
(few seconds) workaround.
Using kernel command line i915.i915_enable_fbc=0 is a permanent workaround.

There were some messages on the Intel-gfx list about this (titled
"Screen corruption regression from 3.0 to 3.1rc4").

Philipp