Subject: Re: [RFC PATCH] mac80211: Add support for hardware ARP query
 filtering
From: Juuso Oikarinen <juuso.oikarinen@nokia.com>
To: ext Johannes Berg <johannes@sipsolutions.net>
Cc: "linux-wireless@vger.kernel.org" <linux-wireless@vger.kernel.org>
In-Reply-To: <1274786743.3635.33.camel@jlt3.sipsolutions.net>
References: <1274773685-11168-1-git-send-email-juuso.oikarinen@nokia.com>
	 <1274778802.3635.30.camel@jlt3.sipsolutions.net>
	 <1274779914.5277.216.camel@wimaxnb.nmp.nokia.com>
	 <1274786743.3635.33.camel@jlt3.sipsolutions.net>
Content-Type: text/plain; charset="UTF-8"
Date: Tue, 25 May 2010 14:34:39 +0300
Message-ID: <1274787279.5277.351.camel@wimaxnb.nmp.nokia.com>
Mime-Version: 1.0
Sender: linux-wireless-owner@vger.kernel.org

On Tue, 2010-05-25 at 13:25 +0200, ext Johannes Berg wrote:
> On Tue, 2010-05-25 at 12:31 +0300, Juuso Oikarinen wrote:
> 
> > > BTW, how will drivers deal with getting this while unassociated? If, for
> > > example, I set an address before associating, you'll get it while
> > > unassociated and not again when associated. Another thing to document --
> > > driver needs to handle that, DHCP is not everything :)
> > 
> > This is actually a good point. The wl1271 driver configures the ARP
> > filter immediately - associated or not - once the chipset is booted up
> > (as above.) This is indeed not very generic.
> > 
> > I will move the set_arp_filter -thingy from _open() to when the we are
> > associated. That way we get the initial IP address (if there happen to
> > be any) from start when the address is actually needed. 
> 
> But then you would still get a callback way before the hardware is even
> started if you set the address before ifup ... mac80211 should probably
> catch that. Or does the networking stack already defer it then?

True again. The call-backs will not get deferred by the stack. That must
also be handled.

> Another thing that came to mind -- how about locking? Will the callback
> have to be atomic? Can you really access idev->ifa_list without any
> locking at all?

AFAIK the callback need not be atomic. I will need to check about the
locking, but we probably do need that.

-Juuso

> > > This is obviously broken when you have multiple virtual interfaces, so
> > > you either need to build a common list of IP addresses, or punt the
> > > problem to the driver and give the callback an ieee80211_vif argument
> > > and clearly document that the driver will have to keep track of it for
> > > each interface.
> > 
> > Good point. I will "punt" (never heard this expression before!) 
> 
> Don't take my word for it ... I heard that once but can't find reference
> to it being a correct use of the word ;)
> 
> johannes
>