Re: NetworkManager and mac80211_hwsim

[Daniel Wagner <wagi monom org>]

On Wed, Feb 18, 2009 at 09:35:59AM +0100, Johannes Berg wrote:
> On Wed, 2009-02-18 at 09:06 +0100, Daniel Wagner wrote:
> 
> > +	if (parent_dev != NULL) {
> > +		hal_device_property_set_string (d, "info.parent", hal_device_get_udi (parent_dev));
> > +		hal_device_property_set_string (d, "net.originating_device", hal_device_get_udi (parent_dev));
> > +	} else {
> > +		hal_device_property_set_string (d, "info.parent", "/org/freedesktop/Hal/devices/computer");
> > +		hal_device_property_set_string (d, "net.originating_device", "/org/freedesktop/Hal/devices/computer");
> > +		hal_device_property_set_string (d, "info.linux.driver", "mac80211_hwsim");
> > +		parent_dev = hal_device_store_find (hald_get_gdl (), "/org/freedesktop/Hal/devices/computer");
> > +		hal_device_property_set_string (parent_dev, "info.linux.driver", "mac80211");
> > +	}
> 
> This is actually not _that_ bad an idea, because it makes sure everybody
> else will properly link up their parent pointer in sysfs. 

That was exactly the problem. HAL creates only a 'correct' dbus tree if the
sysfs parent pointers exists. The patch just fakes the parent 'computer'.
This 'computer' node holds all parent-less nodes (eg alsa_sequencer, alsa_timer).
One further problem is that the 'computer' node doesn't have a 'info.linux.driver'
property which NM wants to read. Setting this property to 'mac80211' is
not correct for all children of this node.

> An alternative
> that doesn't guarantee that would be to check "is it in the ieee80211
> class".

At the moment NM wants a proper parent->child construct to work. Adding something like
/org/freedesktop/Hal/devices/virtual/mac80211_hwsim as parent for the child
would propably be a solution. As I said I was not really ready for fixing
this up. But if this is something worth to fix I'll do it.

daniel