Bug 76574 - Screen freeze during use of chrome/chromium with acceleration and i915/Haswell
Summary: Screen freeze during use of chrome/chromium with acceleration and i915/Haswell
Status: CLOSED DUPLICATE of bug 77207
Alias: None
Product: DRI
Classification: Unclassified
Component: DRM/Intel (show other bugs)
Version: unspecified
Hardware: x86-64 (AMD64) Linux (All)
: medium normal
Assignee: Ben Widawsky
QA Contact: Intel GFX Bugs mailing list
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2014-03-24 22:05 UTC by sibrus
Modified: 2017-07-24 22:55 UTC (History)
2 users (show)

See Also:
i915 platform:
i915 features:


Attachments
Output of dmesg immediately following this error (64.29 KB, text/plain)
2014-03-24 22:05 UTC, sibrus
no flags Details
Output of /sys/class/drm/card0/error which was referenced in the dmesg output (gzipped) (514.75 KB, application/gzip)
2014-03-24 22:07 UTC, sibrus
no flags Details
Output of dmesg, for 2014-03-28-trusty (49.68 KB, text/plain)
2014-03-29 01:03 UTC, sibrus
no flags Details
Output of /sys/class/drm/card0/error (gzipped), 2014-03-28-trusty (508.84 KB, application/gzip)
2014-03-29 01:03 UTC, sibrus
no flags Details
dmesg from drm-intel-nightly with patches (66.38 KB, text/plain)
2014-04-02 16:48 UTC, sibrus
no flags Details
/sys/class/drm/card0/error output, gzipped, from drm-intel-nightly with patch (347.30 KB, application/gzip)
2014-04-02 16:49 UTC, sibrus
no flags Details

Description sibrus 2014-03-24 22:05:51 UTC
Created attachment 96324 [details]
Output of dmesg immediately following this error

My hardware is an Acer C720 (Intel graphics). The symptoms of the problem is the display completely stops updating, except the mouse cursor continues to move. Clicking and any keypresses have no effect. The system remains in this state for about 15-60 seconds, then it recovers and goes back to normal.

The problem most commonly happens when using chrome or chromium to do any kind of intesive hardware-accelerated rendering. (Hardware accelerated rendering is enabled in chrome/chromium.) Typically a youtube video or and HTML5 game with lots of animations. The problem tends to happen during periods of high CPU usage.

xorg version is 7.7+1ubuntu8

OS is ubuntu 14.04 (amd64)
Comment 1 sibrus 2014-03-24 22:07:13 UTC
Created attachment 96325 [details]
Output of /sys/class/drm/card0/error which was referenced in the dmesg output (gzipped)
Comment 2 Ben Widawsky 2014-03-25 03:57:37 UTC
Would it be possible to try our latest -nightly repository?
http://cgit.freedesktop.org/drm-intel/log/?h=drm-intel-nightly
Comment 3 Daniel Vetter 2014-03-26 19:17:42 UTC
We need a new error state from latest kernels, which also dumps the context object. Which is where the issue seems to happen.
Comment 4 Chris Wilson 2014-03-26 19:19:36 UTC
I predict that this has the form where the context object isn't overwritten (at least not in the first page), but it is dying in the context load (that is my guess).
Comment 5 sibrus 2014-03-26 23:55:23 UTC
I can give it a shot, it may be a few days until I have a free evening to do it. Are there any instructions for compiling/installing it?

Also am I just compiling the driver or do I need to do xorg or anything else?
Comment 6 Daniel Vetter 2014-03-27 08:39:56 UTC
It's the kernel, but nothing else required (it's just a drop-in replacement). Distro's usually have excellent howtos for building random kernels from git and installing them properly, some like Ubuntu even have autobuilders for drm-intel-nightly (called drm-intel-experimental iirc).
Comment 7 Christopher M. Penalver 2014-03-27 10:20:54 UTC
sibrus@gmail.com, you would wnat to follow install instructions from https://wiki.ubuntu.com/Kernel/MainlineBuilds in installing the latest from http://kernel.ubuntu.com/~kernel-ppa/mainline/drm-intel-nightly/ .
Comment 8 sibrus 2014-03-29 01:02:13 UTC
I tested this with the latest kernel available from http://kernel.ubuntu.com/~kernel-ppa/mainline/drm-intel-nightly/ (2014-03-28-trusty)

I was able to reproduce the error, or at least a similar error. The behavior upon encoutering the error was a little different. The screen froze once again, however upon unfreezing instead of fully recovering, all the tabs of chrome wouldn't render anything. After restarting chrome it behaved normally again.
Comment 9 sibrus 2014-03-29 01:03:01 UTC
Created attachment 96575 [details]
Output of dmesg, for 2014-03-28-trusty
Comment 10 sibrus 2014-03-29 01:03:34 UTC
Created attachment 96576 [details]
Output of /sys/class/drm/card0/error (gzipped), 2014-03-28-trusty
Comment 11 Chris Wilson 2014-03-29 11:07:17 UTC
Right the context (first page of, at least) looks intact, but it has the hallmarks of the same context restore bug as the original.
Comment 12 Ben Widawsky 2014-03-31 17:34:05 UTC
Care to try my patch from: https://bugs.freedesktop.org/show_bug.cgi?id=75724
Comment 13 sibrus 2014-03-31 22:47:17 UTC
Hi Ben,

I'm willing to give the patch a shot. I saw two patches attached to the bug you linked me. Do I use both or just the latest one?
Comment 14 sibrus 2014-04-02 16:47:49 UTC
I tried the latest drm-intel-nightly. I downloaded the following commit: http://cgit.freedesktop.org/drm-intel/commit/?h=drm-intel-nightly&id=346aad79969a02f2b9fc77bdade02c7413f42ea9

Then I applied both patches that I found in this bug report: https://bugs.freedesktop.org/show_bug.cgi?id=75724

As soon as I booted with the compiled kernel, I got a few screen freezes, then unity appeared to crash as I lost the unity bar and all window decorations. I did capture the dmesg output and the gfx crash log, which are attached below.
Comment 15 sibrus 2014-04-02 16:48:22 UTC
Created attachment 96794 [details]
dmesg from drm-intel-nightly with patches
Comment 16 sibrus 2014-04-02 16:49:03 UTC
Created attachment 96795 [details]
/sys/class/drm/card0/error output, gzipped, from drm-intel-nightly with patch
Comment 17 Chris Wilson 2014-04-30 10:58:18 UTC

*** This bug has been marked as a duplicate of bug 77207 ***


Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.