Outright stubbornness caused me to persist until I found my rogue
piece of assembly language (allegedly fully tested, cough, cough) that
was decrementing, blocking, and incrementing the "waiting" field of
the semaphore instead of where it was supposed to.
Kernel semaphores are so damn robust to this sort of shenanigans 8),
that it was especially hard to get something wrong to happen.
Well, at least I understand both the old and new semaphore code pretty
well now. 8)
Sorry to waste anybody's time.
-- Perry Wagle (wagle@cse.ogi.edu)