If I recall correctly, Tom pointed out a while back that the comment
justifying the lockless read of the VM bit was not correct (or at least
not complete).
I rewrote it, but it was part of a patch that was not accepted. Attached
is the comment patch only.
Regards,
Jeff Davis