Re: Use case for TASKS_RCU

[Steven Rostedt]

On Mon, 22 May 2017 17:00:36 -0700
"Paul E. McKenney" <paulmck@xxxxxxxxxxxxxxxxxx> wrote:

> On Fri, May 19, 2017 at 12:06:09PM -0700, Paul E. McKenney wrote:
> > On Fri, May 19, 2017 at 10:23:31AM -0400, Steven Rostedt wrote:  
> > > On Fri, 19 May 2017 10:04:21 -0400
> > > Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> > >   
> > > > On Fri, 19 May 2017 06:35:50 -0700
> > > > "Paul E. McKenney" <paulmck@xxxxxxxxxxxxxxxxxx> wrote:
> > > >   
> > > > > Simpler would be better!
> > > > > 
> > > > > However, is it really guaranteed that one SCHED_IDLE thread
> > > > > cannot preempt another?  If not, then the trampoline-freeing
> > > > > SCHED_IDLE thread might preempt some other SCHED_IDLE thread
> > > > > in the middle of a trampoline. I am not seeing anything that
> > > > > prevents such preemption, but it is rather early local time,
> > > > > so I could easily be missing something.
> > > > > 
> > > > > However, if SCHED_IDLE threads cannot preempt other threads,
> > > > > even other SCHED_IDLE threads, then your approach sounds
> > > > > quite promising to me.
> > > > > 
> > > > > Steve, Peter, thoughts?    
> > > > 
> > > > SCHED_IDLE is the swapper task. There's one on each CPU, and
> > > > they don't migrate. And they only get called when there's no
> > > > other task running.  
> > > 
> > > Peter just "schooled" me on IRC. I stand corrected (and he may
> > > respond to this email too). I guess any task can become
> > > SCHED_IDLE.
> > > 
> > > But that just makes this an even less likely option for
> > > synchronize_rcu_tasks().  
> > 
> > Hmmm...  The goal is to make sure that any task that was preempted
> > or running at a given point in time passes through a voluntary
> > context switch (or userspace execution, or, ...).
> > 
> > What is the simplest way to get this job done?  To Ingo's point, I
> > bet that there is a simpler way than the current TASKS_RCU
> > implementation.
> > 
> > Ingo, if I make it fit into 100 lines of code, would you be OK with
> > it? I probably need a one-line hook at task-creation time and
> > another at task-exit time, if that makes a difference.  
> 
> And please see below for such a patch, which does add (just barely)
> fewer than 100 lines net.
> 
> Unfortunately, it does not work, as I should have known ahead of time
> from the dyntick-idle experience.  Not all context switches go through
> context_switch().  :-/
> 
> I believe this is fixable, more or less like dyntick-idle's
> half-interrupts were fixable, but it will likely be a few days.  Not
> clear whether the result will be simpler than current TASKS_RCU, but
> there is only one way to find out.  ;-)
> 


Hi Paul,

I'm currently traveling and don't have the energy to look at code at
the moment. Hopefully, I can look at this more on Thursday or Friday.

Thanks!

-- Steve