Most important is the view of a filesystem developer.
When I did a bit of filesystem hacking, I had no idea
what race conditions I might be creating. I was not
even sure what functions could block, and under what
conditions that would be safe.
There are a limited number of kernel experts, and their
time is limited. I think that a bit of time spent with
documentation will save many hours (weeks?) of time
spent fixing buggy code the rest of us could write.
Sufficient documentation would let a new kernel hacker
write a new filesystem completely from scratch, without
reference to potentially unrelated filesystems.
(specific question: how can a file driver in /proc know
what symlink (if any) was used to reach the file?)