260.xx drivers break Java2D OpenGL pipeline
 

I have been using the Java2D OpenGL pipeline for some time with Java 1.6 on 64-bit linux on several machines with 173.xx, 180.xx, and 195.xx versions of NVIDIA drivers. I haven't had any problems until now.

A few days ago I began trying the new 260 series of drivers. I have tried 260.19.06 and 260.19.12 on 2 machines running Debian 5.0 and Ubuntu 10.10. One machine has an 8400 GS, the other has a Quadro FX 550. The Java2D OpenGL pipeline worked flawlessly on both machines with older 173/180/195 drivers. I am using Sun Java 1.6.0_22 64-bit. Both machines are running 64-bit kernels.

With the 260.xx drivers, on both machines I see the following problems:

1. If I just set -Dsun.java2d.opengl=true, I see all swing frames are empty - no widgets are drawn (see empty_java2d_opengl_window).

2. If I set -Dsun.java2d.opengl=true and -Dsun.java2d.opengl.fbobject=false (disable frame buffer objects) this works around problem #1 and widgets are displayed (although this has known performance impacts)

3. In either case #1 or #2 (sun.java2d.opengl=true and sun.java2d.opengl.fbobject is true or false), when I close the Java GUI application the X server almost always (but not 100% of the time) crashes with a segmentation fault that involves nvidia_drv.so (captured from /var/log/Xorg.0.log):

Backtrace:
[ 770.382] 0: /usr/bin/X (xorg_backtrace+0x28) [0x4a0fa8]
[ 770.382] 1: /usr/bin/X (0x400000+0x60fcd) [0x460fcd]
[ 770.382] 2: /lib/libpthread.so.0 (0x7f7d47901000+0xfb40) [0x7f7d47910b40]
[ 770.382] 3: /usr/lib/xorg/extra-modules/nvidia_drv.so (0x7f7d42279000+0x3a3a7c) [0x7f7d4261ca7c]
[ 770.382] 4: /usr/bin/X (FreeClientResources+0xef) [0x44bc5f]
[ 770.382] 5: /usr/bin/X (CloseDownClient+0x5c) [0x427c0c]
[ 770.382] 6: /usr/bin/X (0x400000+0x2c3a6) [0x42c3a6]
[ 770.382] 7: /usr/bin/X (0x400000+0x2184b) [0x42184b]
[ 770.382] 8: /lib/libc.so.6 (__libc_start_main+0xfe) [0x7f7d4686cd8e]
[ 770.382] 9: /usr/bin/X (0x400000+0x213d9) [0x4213d9]
[ 770.382] Segmentation fault at address (nil)
[ 770.382]
Caught signal 11 (Segmentation fault). Server aborting


I have a workaround for now in using an older 173/180/195 series of drivers, but I am concerned about the future of the Java2D OpenGL pipeline if it is broken in the current release.

Re: 260.xx drivers break Java2D OpenGL pipeline
 

I can confirm this bug.
256.53 - works
260.xx - Xserver crashing

Re: 260.xx drivers break Java2D OpenGL pipeline
 

I can also confirm that the Java2D OpenGL pipeline doesn't work correctly. Tested Swing-based Java-app with JDK6u22 (Oracle), Ubuntu 10.04 x86-64 (Xserver 1.7, kernel 2.6.32) and Nvidia driver 260.19.12 (NVS 3100M).

Re: 260.xx drivers break Java2D OpenGL pipeline
 

KDE, oyvind - thanks for confirming this bug.

NVIDIA folks (AaronP, others) - it seems this bug is repeatable by multiple people. Can a bug be opened against the NVIDIA 260 drivers for this? What is the timeframe for getting something like this fixed?

Re: 260.xx drivers break Java2D OpenGL pipeline
 

Thanks for reporting this.

Can you please attach a sample app that demonstrates this?

Re: 260.xx drivers break Java2D OpenGL pipeline
 

I believe any Java2D application that tries to use the OpenGL pipeline will show the problems in my original post.

I am attaching the small example I used to take the screenshots in the original post. The code and compiled .class files are in java-test.zip.

Re: 260.xx drivers break Java2D OpenGL pipeline
 

Thanks. I've confirmed the missing widget problem when setting -Dsun.java2d.opengl=true, but haven't seen the X crash yet.

Can you attach an nvidia-bug-report.log from a system after an X crash, so I can try matching your hardware/software configuration more closely?

Re: 260.xx drivers break Java2D OpenGL pipeline
 

Attaching nvidia-bug-report.log.gz. This is from the Debian 5 machine with a Quadro FX 550.

If it's helpful I also have access to an Ubuntu 10.10 machine with a GeForce 8400 GS where I have seen the same problem - let me know if you'd like data from that box also.

Thanks,
Aaron

Re: 260.xx drivers break Java2D OpenGL pipeline
 

danix - Just wondering what the next step is for this. Can a bug/bugs be opened against the Nvidia drivers for these problems? If you need any more data from me please let me know. Thanks.

Re: 260.xx drivers break Java2D OpenGL pipeline
 

I did not get any X crashes either. Tested the Netbeans IDE with OpenGL pipline. But all widgets were gone (only the splash window was drawn correctly). [Specs at http://www.nvnews.net/vbulletin/show...9&postcount=3]

Re: 260.xx drivers break Java2D OpenGL pipeline
 

I've noticed that setting __GL_SINGLE_THREADED=1 causes a segfault in libGL instead of just missing widgets.

Re: 260.xx drivers break Java2D OpenGL pipeline
 

I have found a reliable way to reproduce the X server crash.

I downloaded the source for OpenJDK 6 and found that whenever -Dsun.java2d.opengl=true, java creates at least one pbuffer at startup in jdk/src/solaris/native/sun/java2d/opengl/GLXGraphicsConfig.c in the GLXGC_InitScratchPbuffer function. If -Dsun.java2d.opengl.fbobject=false, java will create additional pbuffers for rendering all 2D widgets.

After some experiments I have come to the conclusion that if a program has an existing pbuffer created with glXCreatePbuffer and the program exits while the pbuffer exists, there is a chance the X server will crash with the segmentation fault in my original post.

I made a small C program to demonstrate this and am attaching it. This C program creates a pbuffer with the same attributes Java uses in GLXGC_InitScratchPbuffer.

To reproduce the crash do the following:

1. Extract crash-test-program.zip.
2. Compile offscreen-gl.c (gcc -Wall offscreen-gl.c -lX11 -lGL -o offscreen-gl)
3. Run run_offscreen_gl.sh This script runs offscreen-gl, sleeps for 1 second, then kills it. It does this in a forever loop.

With the 260.19.06 and 260.19.12 drivers, running run_offscreen_gl.sh will eventually cause X to crash with the same segmentation fault that was in my original post. It varies how many iterations of running offscreen-gl this takes - sometimes X crashes the first time offscreen-gl is killed, sometimes it takes several iterations.

I tried this program with version 173.14.28 of the drivers also - I let it run for about 30 minutes with no X server crashes.