On Thursday, 21 of February 2008, Alan Stern wrote:
quoted text
> On Thu, 21 Feb 2008, Rafael J. Wysocki wrote:
> 
> > > > +bool in_suspend_context(void)
> > > > +{
> > > > +	bool result;
> > > > +
> > > > +	mutex_lock(&suspending_task_mtx);
> > > > +	result = (suspending_task == current);
> > > > +	mutex_unlock(&suspending_task_mtx);
> > > > +	return result;
> > > > +}
> > > 
> > > If suspending_task == current then you are guaranteed to be serialized, 
> > > because everything a single task does is serial.
> > 
> > As I said before (but that doesn't seem to reach the list, so I'm repeating),
> > this is to protect other tasks from reading an inconsistent value of
> > suspending_task in case they attempt to remove a device concurrently with
> > respect to us.
> > 
> > While this is not likely to happen right now, because of the freezer, it may
> > very well happen when the freezer is finally removed.
> 
> Sorry, I don't understand.  Are you worried that process A might set
> suspending_task = A but then process B might still see suspending_task
> == NULL?  Or that A might set suspend_task = NULL but then B might
> still see suspending_task == A?
> 
> Neither one will cause any problem, since the only case that matters is
> when B sees suspending_task == B -- and that can happen if and only if
> B was the last process to set suspending_task.
> 
> In fact, you might as well get rid of the set_suspending_task() routine 
> entirely and just put the assignments inline.

OK, I will.

quoted text
> > --- linux-2.6.orig/drivers/base/core.c
> > +++ linux-2.6/drivers/base/core.c
> > @@ -929,6 +929,11 @@ void device_del(struct device *dev)
> >  	struct device *parent = dev->parent;
> >  	struct class_interface *class_intf;
> >  
> > +	if (in_suspend_context()) {
> > +		get_device(dev);
> 
> Where is this get_device() undone?  Shouldn't there be an extra 
> put_device() added to unregister_dropped_devices()?

No, I don't think so, because unregister_dropped_devices() calls
device_unregister() that does the put_device() eventually.

If we are called by device_unregister(), the get_device() is needed to balance
the put_device() that will be called by device_unregister() after we return.

OTOH, if we are called directly, then we need to balance the put_device()
that will be done by device_unregister() called from
unregister_dropped_devices().

I hope I didn't miss anything.

quoted text
> > +		device_pm_schedule_removal(dev);
> > +		return;
> > +	}
> >  	device_pm_remove(dev);
> >  	if (parent)
> >  		klist_del(&dev->knode_parent);
> 
> And now the change to device_destroy() isn't needed at all.

No, it's not.  Didn't I remove it?  I thought I did.

Thanks,
Rafael
--
unsubscribe notice
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/