Bugzilla – Bug 52382
[ivb gt1] Severe image corruption and GPU Hang, too many PS threads
Last modified: 2012-10-08 08:27:34 UTC
Created attachment 64520 [details]
Ubuntu 12.04 login screen
I'm getting Severe graphics corruption on my system. Graphics are fine in Windows 7. I have had the motherboard and CPU replaced, the issue still exists. Memory tests fine.
1) The screen is covered in small wrong-color squares, almost like a corrupt JPG. Some screen elements render fine (e.g: the twm menu) while others are unreadable. Screen elements are not always drawn over when they go away. See the pictures attached for examples:
2) Driver hangs on any 3d type activity. glxgears, for example, will cause the screen to basicly freeze. The driver will then reset (as shown in dmesg). xorg-edgers drivers do not appear to freeze, but glxgears exits with "intel_do_flush_locked failed: Input/output error". glxgears registers 0.163fps or lower.
3) The image seems to display properly using the VESA driver. (Boot gentoo live-dvd with -nofb option)
4) With ubuntu and the xorg-edgers PPA, the corruption is present using both the "uxa" and "sna" AccelMethod options in xorg.conf
5) Sometimes, switching to a VT and back will clean up the image for a moment (i.e. until something moves) Also, after glxgears crashed with the error above, most applications were readable. However, redraw problems were still rampant, and trying to run glxgears again gives the do_flush_locked error above
I have confirmed these issues are present when booting the following distros:
Linux Mint 13 (64bit) Live-dvd
Ubuntu 12.04 (64bit) Install CD and Standard install
Gentoo 64bit Live-DVD 12.1 (regular boot, -nofb boots into vesa mode)
Ubuntu 12.04 with xorg-edgers PPA (As of July 22, 2012)
CPU: Intel i5-3470
Chipset: Intel Z77 (M/B: Asus P8Z77-V, bios rev 1205)
Memory: 2x 8GB Patriot G3 DDR3-12800 1600mhz
Monitor: VGA connected CRT or HDMI connected LCD
Currently Running kernel:
Linux Hoita 3.5.0-5-generic #5-Ubuntu SMP Wed Jul 18 07:35:23 UTC 2012 x86_64 x86_64 x86_64 GNU/Linux
Installed packages (Up-to-date with xorg-edgers-ppa as of July22, 2012):
Created attachment 64521 [details]
Picture of screen artifacts (driconf open in twm)
Created attachment 64522 [details]
Created attachment 64523 [details]
Created attachment 64524 [details]
Listing of installed packages and versions
Created attachment 64525 [details]
Created attachment 64526 [details]
Created attachment 64527 [details]
i915 error state
Created attachment 64528 [details]
lspci -vvv output
Created attachment 64529 [details]
Turns out that the GPU is hanging with the latest xorg-edgers packages, same as before (see the dmesg output). However, the monitor is not resetting like it was (as if the resolution had changed) which is why I had thought otherwise.
We have obviously angered the hw gods here.
Can you try sacrificing some rc6 to see if that appeases them? Please append i915.i915_enable_rc6=0 to your grub boot parameters and see if that makes a difference.
Adding the "i915.i915_enable_rc6=0" command to the kernel boot parameters did not make any change that I can see.
If it would help, I'm willing to checkout the required drivers and compile them with whatever debugging turned on you would like.
Just let me know, I'd really like to see this resolved.
Do you have any overclocking settings in your BIOS?
There are some, but none of them seem to make any difference as far as I can tell.
The Bios has a "Normal" setting in basic mode, which is supposed to disable all the overclocking features. It was the first thing I tried when I got the new motherboard and noticed the problem. In advanced mode, it gives you a LOT more overclocking options, but I'm not into overclocking, so I haven't used many.
Right now, almost everything is set to factory default except the RAM. It's set to 1600MHz, which is what the RAM is rated for. Previously, it was 1333MHz, and the problem still existed.
I have not tried fiddling with any of the more advanced settings. Right now, mostly they're all set to auto.
If there is any particular change you would like me to make, please let me know.
Can you please do a lspci -n? The question of the hour is whether this a 0x0152 (IvyBridge desktop GT1). I have two other bug reports for that specific chip, could this be a third?
I pushed a potential fix to http://cgit.freedesktop.org/~ickle/xf86-video-intel/log/?h=ivb-gt1 if you are able to test, that would be fantastic.
Created attachment 64746 [details]
Output from lspci -n
This is the output of lspci -n as requested
Thanks, so three very similar deaths, each on a 0x0152.
Created attachment 64751 [details]
Logs from running driver snapshot 4a7334eb... from git
These are the log files generated by running the driver version found here:
Results: This _appears_ to fix the "corrupted jpg" look of the graphics on the initial login screen, and when running TWM.
However, when launching Unity (logging in), the computer crashes HARD. Previously, I could switch to a VT, and kill X. Now, I cannot. (I needed to ssh into the system to get these logs)
Also, glxgears crashes without showing anything.
I believe the ddx portion of this to be fixed with:
Author: Chris Wilson <firstname.lastname@example.org>
Date: Thu Jul 26 10:50:31 2012 +0100
Reduce maximum thread count for IVB GT1 to avoid spontaneous combustion
Somewhere along the way it seems that IVB GT1 was reduced to only allow
a maximum of 48 threads, as revealed in the lastest bspecs.
Signed-off-by: Chris Wilson <email@example.com>
Note that a corresponding patch for mesa is also required.
*** Bug 52442 has been marked as a duplicate of this bug. ***
*** Bug 52473 has been marked as a duplicate of this bug. ***
Author: Eric Anholt <firstname.lastname@example.org>
Date: Fri Jul 27 11:34:07 2012 -0700
i965/gen7: Reduce GT1 WM thread count according to updated BSpec.
Acked-by: Kenneth Graunke <email@example.com>
also pushed to 8.0.