Sometimes I got this in the kernel logs:
[drm] PGRAPH_ERROR - nSource: LIMIT_ZETA, nStatus: PROTECTION_FAULT
[drm] PGRAPH_ERROR - Ch 2/2 Class 0x009f Mthd 0x0308 Data 0x00000007:0x011c0130
I checked every calls in DDX driver that set data to method 0x308 for class 0x9f (NV12_IMAGE_BLIT), and did not find anything that may cause the error. It happens either at start time (heavy bus usage, to load everything from disk), when I move a big window (opaque mode), or sometimes more randomly.
I notice some big lags just before it triggers, and X crashes with DMA queue hang error. It's like the fifo data has been overwritten. Enabling NOUVEAU_DMA_DEBUG in DDX did not show anything wrong.
Card is 0x10de:0x0309, NV30GL
Still seeing this?
It does not happen anymore since I bought a new PC to use the NV30.
Maybe it was simply the old motherboard that could not handle it, but I suppose the Nvidia blob driver has some checks to avoid this.
I think you can close this bug now, but it may still happen on old boxes.
You could have closed this as a hardware failure yourself ;-)