Since upgrade to Mesa 7.8.1, xf86-video-intel 2.11.0 and xorg 1.8.0, linux 126.96.36.199 I'm experiencing hard lock-ups on my machine. The PC is quite dead, I can't login over ssh or do some sysrq magic. To reproduce it I don't have to cause any activity, running compiz + gnome, leaving it alone for few minutes seems to be sufficient to trigger it.
There is nothing to see in system logs (xorg, kernel, etc.), even running with nmi_watchdog enabled does not help.
To workaround the problem, I have to remove i965_dri.so, without 3d acceleration, everything runs like before upgrading to above (experimental?) configuration.
Similar bug report can be found in gentoo bugzilla:
# grep Chipset /var/log/Xorg.0.log
[ 26.227] (--) PCI:*(0:0:2:0) 8086:2e22:8086:5002 Intel Corporation 4 Series Chipset Integrated Graphics Controller rev 3, Mem @ 0xd0000000/4194304, 0xc0000000/268435456, I/O @ 0x0000f1c0/8
[ 26.227] (--) PCI: (0:0:2:1) 8086:2e23:8086:5002 Intel Corporation 4 Series Chipset Integrated Graphics Controller rev 3, Mem @ 0xd0400000/1048576
[ 26.342] (II) intel: Driver for Intel Integrated Graphics Chipsets: i810,
[ 26.377] (II) intel(0): Integrated Graphics Chipset: Intel(R) G45/G43
[ 26.377] (--) intel(0): Chipset: "G45/G43"
Downgrading intel driver to version 2.10.x solves the problem.
Looks like I am not alone:
I also experience this exact issue as well.
Disabling page flipping solves the issue for me.
See http://lists.fedoraproject.org/pipermail/devel/2010-May/135429.html for details.
Dear Intel developers. Are you still there?
(In reply to comment #4)
> Disabling page flipping solves the issue for me.
That's very useful information, thanks! I've retitled the bug and assigned to Jesse.
> Dear Intel developers. Are you still there?
Sorry for the delayed reply.
Trying to reproduce this now.
I've verified both the vbltest and modetest programs from libdrm work as expected on my GM45, can you test those and make sure they work correctly for you too?
$ git clone git://git.freedesktop.org/git/mesa/drm libdrm
$ cd libdrm
$ ./autogen --prefix=/opt/gfx-test
$ cd tests/vbltest
$ vbltest # pass "-s" if you're testing a second head or laptop screen
$ cd ../modetest
$ modetest # this will dump the modes and connectors
$ modetest -s 7:1440x900 -v # try page flipping on connector 7 at 1440x900
You'll need to adjust the modetest command above for your platform.
Assuming those work, the problem is likely in the X or GL stack somewhere. I'll test that next as well.
Oh, and both vbltest and modetest -v will report frequencies. Those frequencies should match your refresh rate. If they don't, we have an interrupt related problem somewhere.
Created attachment 35994 [details] [review]
Fix exchange validity check
Seems to flip ok with this patch and current X & xf86-video-intel master.
Spoke too soon; with compiz I'm seeing corruption now, but no hang.
After updating Mesa, flipping appears to work with the patch from comment #9. Can you confirm?
This issue is not exclusive to the gm45. The link in comment two refers to a 965. I'm also using a 965 and have seemingly random hard lockups. Should I open a separate bug for the 965?
Yes, please open a new bug if you're seeing hangs with the latest bits. See http://intellinuxgraphics.org/how_to_report_bug.html for a list of the info we're looking for when filing a bug.