Surely there are other devices out there that are sensitive to
unexpected reads?
Oh yeah, I just checked out the way things _really_ work... hmmm,
maybe we need a vm flag called "VM_DEVICE" which can exist at the same
time as the VM_LOCKED flag and inform the mmap() code you mention not
to try to force the pages in because "they already are in".
I bet this nuance slows down the X server startup a bit.
Later,
David S. Miller
davem@caip.rutgers.edu