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

[Bill Huey (hui) <billh@gnuppy.monkey.org>]

On Thu, Sep 21, 2006 at 09:29:08AM +0200, Ingo Molnar wrote:
> * Bill Huey <billh@gnuppy.monkey.org> wrote:
> > I overloaded another reaping thread that was doing largely similar 
> > functionality in that it was also reaping, so I don't think it's that 
> > bad. I did it from a cleanliness point of view with the code tree. 
> > It's the "desched_thread" in fork.c that I'm using. It seems to be the 
> > right thing to do. I'm sure Esben will follow up on this.
> 
> the reason why i added desched_thread was not because it's "more right" 
> to do this from a separate context, but simply because the resource 

I only did that because I saw it there and I assumed it the was the correct
thing to use and that's why I used it.

> freed by it is not being freed via RCU by the upstream kernel. If that 
> resource (mm_struct) were freed by RCU we'd have its rt-friendly 
> reapdown "for free" and no desched_thread would be needed at all.

Well, it's difficult to say. I can't say which is the best method. If the
upstream kernel used RCU function in a task allocation or task struct reading
in the first place then call_rcu() would be a clear choice. However, I didn't
see it used in that way (I could be wrong) so I use the next closest thing that
seems reasonable which is the thread desched_thread(). It use it to avoid
overloading the sematics of call_rcu() to be anything other than a pure RCU
callback. I suggest talking to Esben an Paul about this to get their view on
the matter.

Either method, call_rcu or desched_thread does the trick outside of the
scheduler path and fixes the problem. It's your choice.

bill