In message <20001011204405.A860@dr-evil.shagadelic.org>,
Jason R Thorpe writes:


>you need to be careful -- if you replace those delays with tsleeps,
>you also need to add lockmgr() locks in the right places to keep
>other processes from getting in there to screw things up.

Thanks again. Got an example I can copy?  I'm looking at the 1.5
branch sys/ic/i82365.c. pcic_delay() uses tsleep() with no locking
that I can see.  (the only use of locks is in the pcic_event_thread).
Or are the locks needed on -current, but not the release branch?



>delay() sort of enforces multi-threaded programming, and we need to
>purge it for that reason anyway, but just so you're aware... :-)

I think that came out backwards -- using delay() means the code has to
run in a single thread -- but I get the picture :-).