Run rcu_gc periodically in a separate kernel thread?

compiling, but unused, primitive rcu

read inode from disk in iget, not ilock

handle out of memory in a few cases

fix some races by having ialloc go thru inode cache

very primitive and unscalable

still need to fix kill

dynamically allocate struct procs

BUT can't figure out how to fix bad kill()/condvar interaction

list of sleepers per condvar

try to be careful not to tread on existing vmas (e.g. stack)

keep the panics for now