VBoxTestOGL segfault regression

[efferre]

I'm using Virtualbox on a Linux host with a Windows guest. If I enable the 3D acceleration the virtual machine is not started because VBoxTestOGL segfaults.

The segfault happens only when I use the latest Virtualbox 4.3.2 and the proprietary ATI driver fglrx, if I use the opensource radeon driver VBoxTestOGL works correctly. The point is that this is a regression because Virtualbox 4.2.18 with the same proprietary ATI driver works fine (VBoxTestOGL doesn't segfault).

I have compiled Virtualbox with the --build-debug flag and tested the VBoxTestOGL in different conditions.

Virtualbox 4.2.18:

• VBoxTestOGL --test=2D >> OK
• VBoxTestOGL --test=3D >> OK

Virtualbox 4.3.2:

• VBoxTestOGL --test=2D >> OK
• VBoxTestOGL --test=3D >> SEGFAULT

I'm attaching the logs of the tests and the backtrace. If you look at the log you can see that the 3D test is completed successfully even with version 4.3.2 but then VBoxTestOGL segaults.

[efferre]

VBoxTestOGL_backtrace_segfault.txt.gz

[efferre]

test2d_4.2.18.log.gz

[efferre]

test3d_4.2.18.log.gz

[efferre]

test2d_4.3.2.log.gz

[efferre]

test3d_4.3.2.log.gz

[efferre]

I have just checked out the source to see the differences but I don't understand the adopted revisioning system. The current SVN has a revision of 49651 while the numbers reported in the attached logs are totally different. The 4.2.18 has a release revision of r88780 while the 4.3.2 has a value of r90405. How do these numbers relate to the subversion revisions? Which are the SVN revisions of the two releases 4.2.18 and 4.3.2?

[Frank Mehnert]

Stable branch builds don't have a corresponding SVN revision in the public repository. Just use the respective tarball, it contains the complete source code.

[efferre]

Can you provide an hint about which part of the code is interested by the VBoxTestOGL?

[efferre]

Replying to frank:

Stable branch builds don't have a corresponding SVN revision in the public repository. Just use the respective tarball, it contains the complete source code.

I'd be good to have some references in the SVN repo to the release versions because it's easier to work with the versioning system to look for a bug.

Anyway I managed to bisect the problem using SVN and the bad commit is revision 44860. Let me know what I can do to further debug the problem.

[efferre]

I have tested the latest SVN revision (49853) and the problem is still there

[aeichner]

Please reopen if still relevant with a recent VirtualBox release.