Re: [Lse-tech] [PATCH 09/11] Task watchers: Add support for per-taskwatchers

[Peter Williams]

Matt Helsley wrote:
> On Tue, 2006-06-20 at 15:28 +1000, Peter Williams wrote:
> > Matt Helsley wrote:
> > > This introduces a second, per-task, blocking notifier chain. The per-task
> > > chain offers watchers the chance to register with a specific task nstead of
> > > all tasks. It also allows the watcher to associate a block of data with the task
> > > by wrapping the notifier block using containerof().
> > >
> > > Both the global, all-tasks chain and the per-task chain are called from the samefunction. The two types of chains share the same set of notification
> > > values, however registration functions and the registered notifier blocks must
> > > be separate.
> > >
> > > These notifiers are only safe if notifier blocks are registered with the current
> > > task while in the context of the current task. This ensures that there are no
> > > races between registration, unregistration, and notification.
> > >
> > > Signed-off-by: Matt Helsley <matthltc@xxxxxxxxxx>
> > > Cc: Jes Sorensen <jes@xxxxxxx>
> > > Cc: Chandra S. Seetharaman <sekharan@xxxxxxxxxx>
> > > Cc: Alan Stern <stern@xxxxxxxxxxxxxxxxxxx>
> > [bits deleted]
> >
> > > Index: linux-2.6.17-rc6-mm2/kernel/sys.c
> > > ===================================================================
> > > --- linux-2.6.17-rc6-mm2.orig/kernel/sys.c
> > > +++ linux-2.6.17-rc6-mm2/kernel/sys.c
> > > @@ -450,13 +450,41 @@ int unregister_task_watcher(struct notif
> > >  	return blocking_notifier_chain_unregister(&task_watchers, nb);
> > >  }
> > >  
> > >  EXPORT_SYMBOL_GPL(unregister_task_watcher);
> > >  
> > > +static inline int notify_per_task_watchers(unsigned int val,
> > > +					   struct task_struct *task)
> > > +{
> > > +	if (get_watch_event(val) != WATCH_TASK_INIT)
> > > +		return raw_notifier_call_chain(&task->notify, val, task);
> > > +	RAW_INIT_NOTIFIER_HEAD(&task->notify);
> > > +	if (task->real_parent)
> > > +		return raw_notifier_call_chain(&task->real_parent->notify,
> > > +		   			       val, task);
> > > +}
> > It's possible for this task to exit without returning a result.
>
> Assuming you meant s/task/function/:

Yes, sorry.

> 	In the common case this will return a result because most tasks have a
> real parent. The only exception should be the init task. However, the
> init task does not "fork" from another task so this function will never
> get called with WATCH_TASK_INIT and the init task.

OK.  But it causes a compiler warning:

/home/peterw/KERNELS/CpuCaps/TW-2.6.17-rc6-mm2/kernel/sys.c: In function 
‘notify_per_task_watchers’:
/home/peterw/KERNELS/CpuCaps/TW-2.6.17-rc6-mm2/kernel/sys.c:464: 
warning: control reaches end of non-void function

> 	This means that if one wants to use per-task watchers to associate data
> and a function call with *every* task, special care will need to be
> taken to register with the init task.

OK.  I think I can safely ignore init.

Peter
--
Peter Williams                                   pwil3058@xxxxxxxxxxxxxx

"Learning, n. The kind of ignorance distinguishing the studious."
 -- Ambrose Bierce
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/