Recovering from NVRM XID errors
Dear NVIDIA supporters, developers, and anyone with the same problem,
After changing my graphics card brand from ATI (due to severe driver problems) to NVIDIA (no problem on my other computer), I am experiencing the well known 'NVRM XID' problems.
The only way I found to recover from the problem is rebooting the computer. My uptime is around two weeks between such events.
My setup is quite complex, three monitors, two graphics cards, two X servers on the NVIDIA card for switching between TV-OUT+main display and dual head, a USB webcam DMA-ing continuously, a self-written OpenGL TV program (+small kernel hack), necessary because the DGA support has been dropped, to watch TV with 0% CPU utilization. The setup is rock solid, except this NVRM XID problem.
The order of PCI cards really matters, with other ordering IRQ sharing problems occurred, and the USB webcam reported DMA problems. I know the south bridge of the mainboard is not the best, but many people with different machines experience the same problem.
What I am asking is not a general fix of the problem. I understand that due to contention of the PCI/AGP bus, etc. DMA timeout errors might occur, resulting in misprogramming the GPU. What I ask for is a modification to the driver by which when the problem occurs
1, there is no 100% CPU consumption
2, after killing the X server the system recovers (resets the GPU), and the X server can be started again without rebooting the machine.
I think this kind of modification is feasible and would make me happy.