On Fri, 2011-03-25 at 09:10 -0400, Steven Rostedt wrote:

One solution is to have this be only done on explicit trylocks. Perhaps
introduce a mutex_trylock_spin()? Then when the developer knows that
this scenario does not exist, they can convert mutex_trylocks() into
this spinning version.

I'm not sure this is even worth it, as I'm looking at the
btfs/extend-tree.c code, this is the main reason to use mutex_trylock().

I guess what you see in your benchmarks is that trylock contention
happens mostly in the non-deadlock scenario. But I bet you have
latencies when it does happen, but the benefit seems to out weigh it in
the results.

I wonder what happens if you run dbench as an RT task.

-- Steve