Re: [HACK] convert i_alloc_sem for direct_io.c craziness!

On Mon, 2007-10-01 at 12:52 -0700, Zach Brown wrote:

Do you have any suggestions for locking constructs that RT would prefer?

Basically, anything that maps to a simple mutex. Anything more complex
gets real messy real quick.

Locks that have non-exclusive states become non-deterministic because an
unbounded number of contexts can be in this state. Hence acquisition of
the exclusive state has unbounded time. Even when limited to a bounded
number, the ramifications to the PI graph will get you a head-ache.

Also, non-owner locks, ie. semaphores (asymetric acquisition vs release
contexts) are unusable because the lack of ownership undermines PI - who
to boost?