To a cursory examination, it looks like it's possible that a number of
   syscalls that touch only process state (particularly read-only ones
   like getcwd, getpid, etc. but also some that change state, like
   chdir, chroot, and a few others) could easily take subsystem locks
   instead of the giant lock.
   
   Is this correct?


getpid() already is does not take the kernel lock.  you are probably right
that more could do this.


.mrg.