Re: [RFC 15/17] cfg80211: add P2P Device abstraction

[Johannes Berg <johannes@sipsolutions.net>]

On Mon, 2012-06-25 at 18:16 +0200, Johannes Berg wrote:
> On Mon, 2012-06-25 at 19:05 +0300, Eliad Peller wrote:
> > On Mon, Jun 25, 2012 at 6:48 PM, Johannes Berg
> > <johannes@sipsolutions.net> wrote:
> > > On Mon, 2012-06-25 at 18:20 +0300, Eliad Peller wrote:
> > >
> > >> > +void cfg80211_unregister_wdev(struct wireless_dev *wdev)
> > >
> > >> > +       case NL80211_IFTYPE_P2P_DEVICE:
> > >> > +               if (wdev->p2p_started) {
> > >> > +                       wdev->p2p_started = false;
> > >> > +                       rdev->opencount--;
> > >> > +               }
> > >> any reason not calling stop_p2p_device() here as well?
> > >
> > > That would be circular? You're supposed to call this when
> > > stop_p2p_device() gets called, or when you need to delete it for some
> > > other reason, usually device unplug or such.
> > >
> > hmm... so where stop_p2p_device is being called when userspace doesn't call it?
> 
> Why should it be called? The stop can either originate from userspace,
> or from the driver in case the device is unplugged etc. In that case,
> the driver is expected to clean up its own status before it calls
> unregister_wdev().
> 
> I guess this is a bit different from the netdevs where we do call the
> relevant stop functions when the interface goes down, and it goes down
> when it is unregistered, etc.
> 
> Then again, if we do call stop_p2p_device here, it's not actually that
> circular because it will only happen the first time around. Hmm.

Wait, it's not actually circular, I'm confusing stop_p2p_device and
del_virtual_intf... So I guess it makes sense and we could just call it
here, and drivers don't need to stop anything etc. I'll change it.

johannes