Re: [PATCH] move put_task_struct() reaping into a thread [Re: 2.6.18-rt1]

["Paul E. McKenney" <paulmck@us.ibm.com>]

On Wed, Sep 27, 2006 at 04:06:59PM +0200, Ingo Molnar wrote:
> 
> * Eric W. Biederman <ebiederm@xmission.com> wrote:
> 
> > I'm still wondering if we can move put_task_struct a little lower in 
> > the logic in the places where it is called, so it isn't called under a 
> > lock, or with preemption disabled.  The only downside I see is that it 
> > might convolute the logic into unreadability.
> 
> well it's all a function of the task reaping logic: right now we in 
> essence complete the reaping from the scheduler, via prev_state == 
> TASK_DEAD. We cannot do it sooner because the task is still in use. I 
> had one other implementation upstream some time ago, which was a 
> single-slot cache for reaped tasks - but that uglified other codepaths 
> because _something_ has to notice that the task has been unscheduled.

I believe that we are way too far into the task-teardown process for
something like synchronize_rcu() to be feasible (not enough of the
task left to be able to sleep!), but thought I should bring up the
possibility on the off-chance that it caused someone to come up with a
better approach.

Another possible approach would be workqueues.  The disadvantages here are
(1) higher overhead (2) workqueues can be delayed for a -long- time in a
realtime environment, which increases vulnerability to memory exhaustion.

Again, hoping this provokes some better ideas...

							Thanx, Paul
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/