HighMemory
GEM memory leak from GLX 1.3/1.4 backport
Recently in Lucid, a major memory leak was found in the X.org server which causes the computer to get slower and slower over time. This is reported as bug 565981. This does not affect cards using proprietary drivers or not using DRI2 because it is specific to the glx module that the open drivers use. Intel will always be affected since DRI2 is used with and without KMS, ATI uses DRI1 without KMS.
One possible solution is to roll back the GLX 1.4 enablement patches, and the patch which caused the memory leak to appear. These GLX patches were produced by RedHat and incorporated into Debian, they were not brought in due to Ubuntu-specific requirements and thus it is believed dropping these patches would not impact any of Lucid's development goals. The one risk to be mindful of is if any userspace applications have come to depend on the newer GLX functionality.
To verify that you have this specific problem, open a terminal and run
grep "object bytes" /sys/kernel/debug/dri/0/gem_objects
If it is a large number (for example: 1GB or higher was possible after 8 hours on intel, a negative value indicates multiple gigabytes of memory lost) then you can confirm that you are experiencing this. If it is not that large and shows no signs that will get that large after 24 hours of uptime then your issues are most likely unrelated to this specific problem.
Even if you are not affected by this, testing of these packages is appreciated and please add your information to the bottom if you do so.
Installing the test packages
Please do the following to upgrade to the proposed X server with the fix:
sudo add-apt-repository ppa:ubuntu-x-swat/x-updates sudo apt-get update sudo apt-get dist-upgrade
After that, reboot and log in.
Testing
Please work with your system normally for a few hours with compiz enabled (at which point the current lucid X.org server should feel really sluggish), as well as check functionaly and performance of:
- Video playback (fullscreen and windowed)
- Games (extremetuxracer, sauerbraten, etc.)
- compiz (quickly change desktops and move windows around)
- Flash movies
- etc.
Please verify that glxinfo | grep "GLX version" says "1.2", not "1.4".
Reverting back to lucid packages
sudo apt-get install ppa-purge sudo ppa-purge ppa:ubuntu-x-swat/x-updates
Feedback
Your name |
Desktop |
Chipset |
Driver (fglrx vs. ati, etc.) |
Bug fixed? |
Any regressions? |
Comments |
|
Martin Pitt |
GNOME |
GMA945 (Dell Latitude D430) |
intel |
Yes |
No |
also fixes long shutdown hang |
|
Martin Pitt |
UNE |
GMA945 (Dell Mini 10) |
intel |
Yes |
No |
|
|
Didier Roche |
GNOME/UNE |
GMA945 (on both netbooks) |
intel |
Yes |
Not yet (after one workday) |
|
|
Ken VanDine |
GNOME |
GM45 |
intel |
Yes |
No |
|
|
Ken VanDine |
UNE |
945GME |
intel |
Yes |
No |
|
|
Sebastien Bacher |
GNOME |
intel 965 |
intel |
(testing) |
testing |
|
|
C de-Avillez |
GNOME |
amd64 |
ati |
(testing) |
testing |
|
|
GNOME |
82G33 |
intel |
(testing) |
(testing) |
|
||
Stefan Lesicnik |
GNOME |
GeForce GTX 260M |
nvidia |
(testing) |
(testing |
Apparently nvidia driver bypasses xserver glx, so not good testcase (works ok though) |
|
Thanh Bui Van |
GNOME |
ATI Radeon |
Intel x86 |
Yes |
No |
|
|
Robert Hooker |
Gnome/UNE |
945GME |
intel/fbdev |
Yes |
No |
|
|
Robert Hooker |
Gnome/UNE |
915GM |
intel/fbdev |
Yes |
No |
|
|
Robert Hooker |
Gnome/UNE |
G33 |
intel/fbdev |
Yes |
No |
|
|
Robert Hooker |
Gnome |
HD3400M |
ati |
Yes |
No |
|
|
Robert Hooker |
Gnome |
SIS 760GX |
sis |
No problem originally |
No |
Updating to the xserver in x-updates on this machine fixes the problem where clutter based apps (such as gnome-games) fail to start when using Software Rasterizer (bug: 561734) |
|
Robert Hooker |
Gnome |
8400M GS |
nouveau |
No problem originally |
No |
As above, clutter apps using swrast (which nouveau uses) are fixed with the xserver in the PPA |
|
Jamie Strandboge |
Gnome |
Radeon 7500 |
ati/KMS/DRI2 |
Yes |
No |
Using the following from the bug shows that gem objects are increasing, but not X memory: pid=`pidof X` ; for t in `seq 1 10`; do eog /usr/share/backgrounds ; echo `grep "object bytes" /sys/kernel/debug/dri/0/gem_objects` `ps ocomm,vsz,rss $pid |grep X`; done |
|
Murat Gunes |
GNOME |
GMA945 |
intel |
Yes |
No |
GEM object count now seems to return to original value after a minute with the test case in the bug |
|
Dave Lentz |
GNOME/KDE |
RV710/RadeonHD4550 |
ati |
yes |
(testing) |
|
|
Gnome |
RV730 PRO/RadeonHD4650 |
ati |
(testing) |
|
|
||
Erik Andersen |
Gnome |
Nvidia Geforce 7300s |
nvidia |
(testing) |
We'll see |
Erik, the nvidia/fglrx drivers are not affected. Please read the description above and then delete this entry. |
|
Colin Williams |
Gnome |
Intel Corporation Mobile 945GM/GMS/GME, 943/940GML Express Integrated Graphics Controller (rev 03) |
intel |
Yes |
unsure |
fixed bug! |
|
Gnome |
Intel Corporation Mobile GM965/GL960 Integrated Graphics Controller (rev 03) |
-intel |
Yes |
No |
Gem object count hovers between 1300 and 1450, grows and shrinks, memory usage hovers around 204398592 object bytes, low of about 161542144 high about 260308992. Played sauerbrate, quadrapassel, and a flash moving full screen |
||
Romain Boissat |
GNOME |
GMA965 |
intel |
Yes |
No |
I was only affected by increasing number of GEM object, memory use seemed quite normal, even after 3 days uptime. PPA packages fixed at least the GEM objects increase |
|
Gary Trakhman |
GNOME |
x4500 |
intel |
Yes |
No |
Seems good |
|
Ed Fardos |
KDE |
nforce460 |
nvidia |
we'll see |
No |
Xorg has blown up for the past four weeks iff GL compositing enabled, xrender is okay: http://sierraglider.com/craiger/P1010006.JPG . I realize we think the bug isn't in the nvidia driver, but that's a screenshot of a blown up xorg and exhausted swap. |
|
Anthony Hook |
GNOME |
NVIDIA 8600M GT |
nouveau |
No |
I don't know |
I was able to have both the objects and object bytes raise until I ran out of memory and x thrashed to it's death. |
|
Anthony Hook |
GNOME |
945GME |
intel |
Yes |
No |
It appeared to work alright |
|
Virgil Brummond |
GNOME |
Intel Series 4 Mobile |
Intel 2.11 |
Yes |
No |
Compiz works as usual. Playing 1080p HD video using gstreamer and 3d games as well. X stays at around 44mb. |
|
Conn O'Griofa |
GNOME |
Radeon IGP 345M |
radeon |
No |
No (but no improvement) |
The problem is not fixed for my R100-based card. After one hour of uptime (watching a Flash video), system has become sluggish and GEM object bytes count swells. |
|
Erick Brunzell |
GNOME |
82945G/GZ |
Intel |
Yes |
No |
|
|
Götz Christ |
KDE |
Intel 82865G rev 02 |
intel |
Yes |
No |
|
|
Chris Halse Rogers |
GNOME/UNE |
GM45 |
intel |
Yes |
No |
|
|
Chris Halse Rogers |
GNOME/UNE |
GeForce 9300M GS |
nouveau |
No problem initially |
No |
|
|
Lex McPhail |
Gnome |
Mobile 945GM/GMS, 943/940GML Express Integrated Graphics Controller |
Intel i915 |
(testing) |
(testing) |
|
|
Brett Howard |
GNOME |
Radeon X300M |
ati |
(Testing In progress) |
|
Slow down was very apparent on this system before reverting |
|
Haggai Eran |
GNOME/UNE |
945GME |
intel |
Yes |
No |
|
|
GNOME |
Intel Corporation Mobile 4 Series Chipset Integrated Graphics Controller (rev 07) |
intel? |
(testing) |
(testing) |
|||
Jan Varho |
GNOME |
Radeon HD 4850 |
ati |
Yes |
No |
2 hours' testing, will update if changes |
|
