__put_task_struct() WARN_ON(!tsk->exit_state)?

[Paul E. McKenney]

Hello!

One of last night's rcutorture TREE03 runs triggered the
__put_task_struct() function's first WARN_ON() complaining that
the tasks's ->exit_state was zero.  This was on mainline merged
into RCU's "dev" branch.  There immediately follows a splat from
refcount_warn_saturate(), so there are two interlaced stack dumps.  :-/

What I do have seems to implicate rt_mutex priority boosting triggered
by RCU's prioritiy boosting.

Left to myself, my next steps are to try to reproduce this.  It happened
once in 150 hours of TREE03, so who knows?

But I figured I should check to see if anyone else is hitting this.

						Thanx, Paul