No, it might well be a real kernel bug. The low-level brw_page() stuff got
changed to take care of _another_ potential bug in 1.99.11, and that made
this older bug re-surface again. I think I nailed them both correctly in
1.99.12 (knock wood). I haven't actually ever seen any of the behaviour
that some people report, but source code debugging certainly indicates
that there could have been a real problem there..
In short, could you take a look at 2.0pre12, and see if the problem is
gone? It should have happened _only_ under very heavy load (although a
slow device like a CD-ROM or a floppy might have made it happen a lot
more easily).
Linus