Bug 51062

Created attachment 63009 [details]
Xorg.0.log

Created attachment 63010 [details]
xrandr --verbose

Created attachment 63011 [details]
.xsession-errors

The bug still present with drm-intel-experimental kernel 3.5.0-994-generic #201207060407.

I do not see any apparent errors in dmesg or Xorg.0.log. Any of the messages in .xsession-errors were there before image disappeared from the displays (X only, consoles works)

compiz (core) - Warn: glXCreatePixmap failed
compiz (core) - Warn: Couldn't bind background pixmap 0x2600001 to texture

(gnome-settings-daemon:1978): color-plugin-WARNING **: failed to reset xrandr-AU Optronics gamma tables: gamma size is zero

(gnome-settings-daemon:1978): color-plugin-WARNING **: failed to reset xrandr-Goldstar Company Ltd-L2000C-162469 gamma tables: gamma size is zero

I'm not yet seeing anything to indicate this is anything more than compiz not redrawing. :|

Chris, do you think that the following error from
the original report is due to compiz as well?

If I try xrandr from the console it reports:
DISPLAY=:0 xrandr --output LVDS1 --auto
xrandr: Configure crtc 1 failed

I can admit it is another bug, unfortunately there are
more than enough of them for "production" system.
Fortunately I have not seen severe hangs for some time.

P.S. Just now in Ubuntu quantal unity+compiz do not want
to draw icons in launcher (left panel) and in the window
appearing on [Alt+Tab]. X server with just an xterm still
become dark after LVDS1<->VGA1 switching.

(In reply to comment #6)
> Chris, do you think that the following error from
> the original report is due to compiz as well?
> 
> If I try xrandr from the console it reports:
> DISPLAY=:0 xrandr --output LVDS1 --auto
> xrandr: Configure crtc 1 failed

No, that's a completely separate bug which I thought was being tracked separately already.

Chris, bad news for you. It is not compiz.
With ubuntu drm-intel-experimental kernel
3.5.0-994-generic #201207270423 SMP Fri Jul 27 08:31:26 UTC 2012 i686 i686 i686 GNU/Linux
I managed to reproduce a kind of this bug just with xterm.
The thing I do not like at all, that this time the issue
has quantum nature (Bad idea with "Quantal" name...).
If I set a timer with script, gathering some data
(dmesg, Xorg.0.log, regdump, i915_error state) the xterm
appears when this script executes.

startx /usr/bin/ck-launch-session xterm
# connect external monitor
xrandr --output VGA1 --auto --output LVDS1 --off
# black screen
# [Ctrl+Alt+F1] [Alt+F7]
# exit from xterm [Ctrl+d] (VGA1 is active, LVDS1 is not)
# disconnect external monitor
startx /usr/bin/ck-launch-session xterm
# connect external monitor
xrandr --output VGA1 --auto --output LVDS1 --off
# black screen
# [Ctrl+Alt+F1] [Alt+F7]
# Xorg log on the screen, no response to keyboard

Stale image, this time vt1.

The only way to bring the system to a sane state I found is
to unplug VGA cable. X server stops correctly on [Ctrl+d]
(in xterm) and new commands can be entered at vt1.

I have no idea how to debug this issue (only with my laptop).
I have a dmesg, but that script somehow restored the image.

Again, that's not the bug you are reporting here. This bug is where we switch back to X and we have the X cursor, xterm responsive but only a black background.

Sorry for the delay. Besides Bug 52140, [Fn+F8]
and other Fn keys do not work ubuntu quantal with
current xorg-edgers ppa.

Chris, I faced this problem simultaneously with
  xrandr: Configure crtc 1 failed
If you believe that this bug is solely related to compiz.
feel free to close it. Then I will report the issue
described in Comment 8 as a new bug.

Yeah, even if we see a modeset failure, the original mode should be restored and the image should not "stick". There's a bug there between X and compiz for sure, and I'd like to keep that separate from addressing the bug that is causing the modeset to fail in the first place.

Finally, I downgraded libgl1-mesa-dri from
8.1~git20120725.cc44aa77-0ubuntu0sarvatt 0
        500 http://ppa.launchpad.net/xorg-edgers/ppa/ubuntu/ quantal/main
to
8.0.4-1ubuntu1 0
        500 http://linux.nsu.ru/ubuntu/ quantal/main i386 Packages
Other mesa libraries are from xorg-edgers (quantal).
So Unity desktop is displayed correctly (at least just after login).
There was a strange issue with Fn keys. It disappeared after
I touched hotkeys setup. Last tests are done with a new user
with clean profile. Anyway sometimes I have to remove "mirror"
tick in display setup.

Kernel is taken from drm-intel-experimental:
3.5.0-994-generic #201207270423 SMP Fri Jul 27 08:31:26 UTC 2012 i686 i686 i686 GNU/Linux

After several display switches by [Fn+F8] desktop image become
shifted and wrapped around. I have not noticed anything suspicious
in dmesg and Xorg.0.log. Although I was not able to switch screens
from the console:
DISPLAY=:0 xrandr --output LVDS1 --auto
xrandr: Configure crtc 1 failed

So compiz bug and driver's one work in companion.

Is it normal that CRTC have numbers 3 and 4 in dmesg,
but xrandr complains about crtc 0 or 1?

I did not managed to get LVDS1 working even after I killed X
and started it again. In response to [Ctrl+Alt+F1] [Alt+F1]
image blinked for a moment and the screen became blank in X.
So the problem lies deeper than in compiz.

Any ideas how to debug the issue? Will intel_reg_dumper help you?

I can add a bit more.

The issue described in Comment 8 was caused by the Bug 52440

It seems that now I always get
DISPLAY=:0 xrandr --output LVDS1 --auto
xrandr: Configure crtc 1 failed
when I trying to switch screens from console.
If I switch back to X, the command works.
Is it normal?

I have noticed the following errors in Xorg.0.log:
[   203.293] (II) intel(0): Modeline "1600x1200"x0.0  162.00  1600 1664 1856 2160  1200 1201 1204 1250 +hsync +vsync (75.0 kHz e)
[   203.301] (--) synaptics: SynPS/2 Synaptics TouchPad: touchpad found
[   216.536] (II) AIGLX: Suspending AIGLX clients for VT switch
[   238.256] (II) intel(0): switch to mode 1280x800 on crtc 3 (pipe 0)
[   238.256] (EE) intel(0): get_fb: failed to add fb: 1600x1200 depth=24, bpp=32, pitch=6656: 13
[   238.274] (II) intel(0): switch to mode 1280x800 on crtc 3 (pipe 0)
[   238.274] (EE) intel(0): get_fb: failed to add fb: 1280x800 depth=24, bpp=32, pitch=5120: 13
[   255.274] (II) Open ACPI successful (/var/run/acpid.socket)
[   255.274] (II) AIGLX: Resuming AIGLX clients after VT switch
[   255.274] (II) intel(0): switch to mode 1280x800 on crtc 3 (pipe 0)
[   255.312] (II) intel(0): EDID vendor "AUO", prod id 17476
[   255.312] (II) intel(0): Printing DDC gathered Modelines:
[   255.312] (II) intel(0): Modeline "1280x800"x0.0   68.94  1280 1301 1333 1408  800 803 808 816 -hsync -vsync (49.0 kHz eP)

If I remember correctly, compiz was working at that time.

I wonder if the "[dix] EventToCore" records mean a bug?
Usually they appears when compiz does not properly redraw
the screens.
(from anothe Xorg.0.log)
[   258.127] (--) synaptics: SynPS/2 Synaptics TouchPad: touchpad found
[   258.755] [dix] EventToCore: Not implemented yet
[   271.388] (II) AIGLX: Suspending AIGLX clients for VT switch

And finally some speculations. It seems that compiz allows more
switches of LVDS1 and VGA1 by [Fn+F8] if I do not switch to
console and back ([Ctrl+Alt+F1], [Alt+F7]) in the meanwhile.

So the stale image upon switching was likely to be the UXA pageflip vs modeset race. (And possibly the similar kernel bug.) So I believe this now to be fixed at last!