Re: [Lse-tech] Re: [ckrm-tech] Re: [PATCH 00/01] Move Exit Connectors

[Matt Helsley <matthltc@us.ibm.com>]

On Fri, 2006-01-13 at 05:38 -0500, Jes Sorensen wrote:
> >>>>> "Matt" == Matt Helsley <matthltc@us.ibm.com> writes:
> 
> Matt> On Thu, 2006-01-12 at 04:51 -0500, Jes Sorensen wrote:
> Matt> Have you looked at Alan Stern's notifier chain fix patch?  Could
> Matt> that be used in task_notify?
> >>  No sorry, do you have a pointer?
> 
> Matt> No problem. Here it is:
> Matt> http://marc.theaimsgroup.com/?l=linux-kernel&m=113407207418475&w=2
> 
> Matt> I think it would be ideal if task_notify could simply be a
> Matt> notifier chain for notifying users of task events/changes.
> 
> Ok, went back and looked at this. I think the core concept is fine,
> but there are details such as having a data pointer associated with
> the notifier block which is too important to leave out. Otherwise we
> have to stick things into the task struct in many cases which is a
> waste of space. I also think it needs to be possible to search the
> list for special slow path uses to avoid us adding excessive amounts
> of callbacks that are only used in one place by one client.

	I agree that having a data pointer associated with the notifier block
is important. It helps us avoid increasing the size of task struct for
each task_notify user and makes modularization of them possible.

	Hmm, yes excessive amounts of callbacks for those used by only one
client could be a problem. My first approach to that problem would be to
have one task_notify list per-event rather than just a single list for
all events. This has ugly space implications.

	More importantly, I don't think it's a problem yet. Until it's a
problem we ought to go with the simpler implementations. When it is a
problem the task_notify interface should insulate the users from such a
change.

> If we can cross-API it for task-group-notifiers then that should be
> fine.
> 
> Cheers,
> Jes

Yup.

Cheers,
	-Matt Helsley