Created attachment 52456 [details]
dmesg.txt is the bug's dmesg information

I can reproduce it with xfce as well, and probably other DEs.

The workaround for me is to go to console (ctrl-alt-f1), and then back to X (ctrl-alt-f7). The login proceeds, but sometimes the WM does not starts.

(In reply to comment #2)
> I can reproduce it with xfce as well, and probably other DEs.
> 
> The workaround for me is to go to console (ctrl-alt-f1), and then back to X
> (ctrl-alt-f7). The login proceeds, but sometimes the WM does not starts.

Then please do this: get into the "bad" state, use intel_reg_dumper. Then go to console and back, and if everything is working, get another intel_reg_dumper. We might want to compare both states...

Khm.. what you define by 'bad' state here, the state of intel_reg_dumper at while the computer is running 'gdm'?

During the login, as soon as we switch to console, it changes the registers, and login proceeds. So there is no diff.

But I think you have a point, I'll write a custom session starting script and run it instead of gnome login to see if there are any changes.

Works for me with my ivb-3-pipe branch...  lemme try with 3D.

Part of the problem might be the server's default config chooser.  I posted a fix to xorg-devel for it (http://lists.freedesktop.org/archives/xorg-devel/2011-October/026280.html), but in the meantime you may need to force gdm to choose a 3 pipe compatible config, e.g.:

Section "Device"
  Identifier "IVB"
  Driver "intel"
  Option "monitor-HDMI1" "hdmi1"
  Option "monitor-HDMI2" "hdmi2"
  Option "monitor-VGA1" "vga"
EndSection

Section "Monitor"
  Identifier "hdmi1"
  Option "PreferredMode" "1280x1024"
  Option "RightOf" "vga"
EndSection

Section "Monitor"
  Identifier "hdmi2"
  Option "PreferredMode" "1280x1024"
  Option "RightOf" "hdmi1"
EndSection

Section "Monitor"
  Identifier "vga"
  Option "PreferredMode" "1920x1200"
EndSection

Section "Screen"
  Identifier "Default Screen"
  Device "IVB display config"
  Monitor "vga"
  DefaultDepth 24
  SubSection "Display"
    Depth 24
    Modes "1920x1200" "1280x1024"
  EndSubSection
EndSection

(In reply to comment #6)
> Part of the problem might be the server's default config chooser.  I posted a
> fix to xorg-devel for it
> (http://lists.freedesktop.org/archives/xorg-devel/2011-October/026280.html),
> but in the meantime you may need to force gdm to choose a 3 pipe compatible
> config, e.g.:
> 
> Section "Device"
>   Identifier "IVB"
>   Driver "intel"
>   Option "monitor-HDMI1" "hdmi1"
>   Option "monitor-HDMI2" "hdmi2"
>   Option "monitor-VGA1" "vga"
> EndSection
> 
> Section "Monitor"
>   Identifier "hdmi1"
>   Option "PreferredMode" "1280x1024"
>   Option "RightOf" "vga"
> EndSection
> 
> Section "Monitor"
>   Identifier "hdmi2"
>   Option "PreferredMode" "1280x1024"
>   Option "RightOf" "hdmi1"
> EndSection
> 
> Section "Monitor"
>   Identifier "vga"
>   Option "PreferredMode" "1920x1200"
> EndSection
> 
> Section "Screen"
>   Identifier "Default Screen"
>   Device "IVB display config"
>   Monitor "vga"
>   DefaultDepth 24
>   SubSection "Display"
>     Depth 24
>     Modes "1920x1200" "1280x1024"
>   EndSubSection
> EndSection

Hi Jesse, with your two patches(one is mentioned above and the other is attached) and the xorg config, the 3 monitors works well while running X.

Created attachment 53657 [details]
xserver-initial-config-match-refresh

*** Bug 43011 has been marked as a duplicate of this bug. ***

X server fix is upstream now, closing.

(In reply to comment #10)
> X server fix is upstream now, closing.
OK,it can work well.

Jesse, 

Is the fix on kernel side made into 3.2 kernel?

Is the xserver fix on master, or already in any release?

I think this is a combination of the core and i915 kernel fixes, but maybe the X fix too...

I want to know if the kernel fixes are in 3.2, so we could remove one dependency (current this bug is marked as blocking two kernel releases)

Closing old verified.