Re: HT action frame code

["Tomas Winkler" <tomasw@gmail.com>]

On Tue, Sep 9, 2008 at 11:20 AM, Johannes Berg
<johannes@sipsolutions.net> wrote:
> On Tue, 2008-09-09 at 11:14 +0300, Tomas Winkler wrote:
>
>> >> The current design assumes that hostapd takes care of all management
>> >> frames, so eyes, these would need to go to hostapd for processing and
>> >> then setup back to mac80211 through some new command..
>> >
>> > Well we can always pick out those action frames in the kernel if we want
>> > to, that's not the problem, is it?
>>
>> We've indeed has kept inside just BA session action frames.
>
> Hmm? I don't think I understand what you're trying to say.

Meaning that other management frames are forward to user space while
BA action frames
are treated inside mac80211.

>
>> > Should the design be changed? It seems that this is more related to the
>> > rate scale algorithm and the "can we support aggregation for this STA"
>> > question, both of which we currently have information about in the
>> > kernel and not hostapd.
>>
>> The decision might come both from RS and hostapd. RS just have more
>> info whether there will be gain from aggregation. When throughput is
>> not high enough it's just an overhead. Though some APs open session
>> immediately uppon association without and RS decision.
>> Our design is that BA can be triggered not only for RS.
>
> Well yeah, but I'm not sure we really want to add API to cfg80211 for
> this. It's not exactly hard to, but right now I don't see how hostapd
> would influence the decision.
>
> Also, I'm not talking about the AP triggering the aggregation session,
> this is entirely done with the rate scaling right now, but about an
> associated STA wanting to start an aggregation session. Aren't
> aggregation sessions always triggered by whoever wants to send? So if a
> STA notices it has lots of upload going on it could want to trigger a BA
> session, which is something we don't currently support afaict.

In iwl-agn-rs.c There is not difference if the peer is STA or AP. So
we support this already.
Thanks
Tomas