Re: [PATCH] cpuset semaphore depth check optimize

[Simon Derr <Simon.Derr@bull.net>]

On Tue, 20 Sep 2005, Robin Holt wrote:

> Paul,
> 
> Can you give a _short_ explanation of why notify_on_release is
> essential?  Could the intent be accomplished with something
> like destroy on exit which then goes through and does the
> remove of shildren and finally removes the cpuset?
> 
> If we can agree on that, then the exit path becomes
> 	if (atomic_dec_and_lock(&current->cpuset.refcount)) {
> 		/* Code to remove children. */
> 	}
> which no longer needs to call a usermode helper and is _FAR_
> better in my personal biased opinion.
> 
> 

The original cpuset patch did not have a notify_on_release feature, it had 
an 'autoclean' feature, that did exactly what you describe.

But the locking was not clean, mostly at the filesystem level. Paul and I 
tried together to fix it, but at some point Paul proposed this solution, 
to use a usermode helper, and it seemed to be the easiest way to get out 
of this locking issue, so we agreed to drop the old 'autoclean' scheme.

(Even if Paul had a hard time convincing me, as I did not like the idea of 
this usermode helper too much).


	Simon.
-
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/