glBindFrameBuffer crash on 290.10/319.17 driver on Quadro FX 570/3700 and GeForce GTX580 in Qt 4.7.3

Note: Also seen on the 290.10 driver on a Quadro FX 3700, and a GeForce GTX580.

The Quadro FX 570 was on a 32 bit RHEL 5, 2.6.18-238.el5PAE
- Intel Xeon E5420.
- Two displays, one being a Samsung Syncmaster 2443 (1920x1200), and the other a Samsung HD TV (appears as a 1920x1080 Syncmaster), on which the application was running. The log attached is from this machine.

The Quadro FX 3700 was running on 32-bit Cent OS 5.6, Kernel: 2.6.18-308.16.1.el5PAE
- Intel Xeon W3520
- Monitor Model #: Dell 2007FP (1600x1200)

A notable portion of a trace is here, from which a SIGSEGV is launched:

#10 0x0726e387 in ?? () from /usr/lib/libGL.so.1
#11 0xb6eeb294 in ?? () from /usr/lib/libnvidia-glcore.so.319.17
#12 0xb6eeb39d in ?? () from /usr/lib/libnvidia-glcore.so.319.17
glBindFramebuffer( GL_FRAMEBUFFER_EXT, frameBufferId_);

The frame buffer in question was allocated at application startup, and this call would have been invoked thousands of times before the problem.

Also, the OpenGL context was available, as a call to map a texture and upload data through a pixel buffer object had just succeeded.

The application is multithreaded, but all OpenGL operations are single threaded. There are multiple contexts, as QT generates a couple more for its widget rendering.

This problem looks similar to one encountered in KDE/Firefox: https://bugs.kde.org/show_bug.cgi?id=279546#c1

Questions:

  1. Is anyone aware of frame buffer object related crashes in the nVidia driver that have been corrected later?

I am aware of this in the 325.15:

“Fixed a bug that could cause OpenGL applications that use Frame Buffer Objects (FBOs) to crash following a mode switch (e.g. changing the resolution of a display or transforming it)”

… but there was no resolution change. Would this apply still? Are there any other problems that might be relevant?

Upgrading drivers is not an easy task on my end because of deployment issues. So if there is a known problem and fix, that can be a rationale to justify the cost.

  1. Can this sort of thing occur because there isn’t enough video memory? Note that only a color texture is bound to the FBO, and it was a 512x512 32 bit texture, but it was preallocated, and had definitely been working.

I have an nvidia log from one of the machines, but I am unsure as to how to share it on the forum.