On Tue, 11 Apr 2017 17:34:47 -0400 Steven Rostedt <[email protected]> wrote:
> On Tue, 11 Apr 2017 17:31:33 -0400 > Steven Rostedt <[email protected]> wrote: > > > The thread gets created when I enable the benchmark tracepoint. It just > > so happens that my test enables *all* tracepoints, which would of > > course include this one as well. > > > > I'll have to look at this code to see why it is getting missed. > > Yep, this thread never goes to sleep, but will call cond_resched() > periodically. This keeps rcu_tasks() from finishing. > > Should I add a direct "schedule()" in there instead of a > cond_resched(), or do you think rcu_tasks should have cond_resched() be > a quiescent state as well? Actually, I believe this found a bug in my trace_event benchmark thread. On a preempt kernel, cond_resched() is a nop and expects to only be preempted. Calling schedule() directly should fix everything. I shouldn't depend on cond_resched() here. -- Steve
