Maybe the odds of it doing so are smaller, but if you think a broken nvidia.ko can't screw up anyone's data I think you're simply mistaken. Say Nvidia's driver

Maybe the odds of it doing so are smaller, but if you think a broken nvidia.ko can't screw up anyone's data I think you're simply mistaken. Say Nvidia's driver has a use-after-free bug: it kmalloc()s a buffer, kfree()s it, then a filesystem kmalloc()s something and gets allocated the same buffer. If nvidia.ko then decides it still wants to use that buffer and writes something into it...kablooie. Unless you start running some microkernel-ish thing with drivers each running in their own distinct address spaces, you're going to have a hard time avoiding this possibility.