On Fri, Jan 18, 2013 at 11:00:11PM +0100, Johannes Berg wrote: > On Thu, 2013-01-17 at 14:52 +0100, Simon Wunderlich wrote: > > > > > + res = drv_start_radar_detection(local, sdata, chandef); > > > > > > If the vif is assigned the channel, why also pass it to the > > > start_radar_detection command? That seems pointless, they can't be > > > different? > > > > > In the initial phase, __nl80211_set_channel() should not be able to set > > the channel (no CAC yet), and will fail for IBSS (to be implemented later) > > generally, at least as it's implemented now. > > > > But what we can do is set the channel from mac80211 and call the driver > > function without the channel argument, as it'll be pointless in this case. > > Is this what you mean? > > Hmm. Why are you talking about __nl80211_set_channel() now? We're now > talking about mac80211, which can and does set the channel (chanctx, > which may fall back to drv_config()) before calling > drv_start_radar_detection(), so I think the latter needs no chandef > argument? > Sorry, I was confused at the point, you are right nl80211_set_channel() has nothing to do with that. ieee80211_vif_use_channel() already sets the channel implicitly (e.g. through ieee80211_hw_config()). > > > Actually that raises another question: If we have "external" radar > > > detection, say by a different NIC, then shouldn't we still ask the > > > driver to start radar detection when using the channel? Or is that > > > implicit, does the driver have to check? > > > > That is a good question, I didn't consider that. In the "simple process" > > we first start radar detection and start the ap on the same channel. > > For external CAC that won't work, of course. What about mac80211 checking > > the channel in start_ap(), and if the channel requires DFS, pass some flag > > to the driver that radar detection should be enabled? > > Yes, that makes sense. But then it would also make sense to remove the > start_radar_detection() callback entirely, and encode all that > information in the channel context/drv_config call? > > If mac80211 gets to be responsible for it, this should totally be > documented in the cfg80211 API though so if a full-mac driver wants to > implement it they know what to do. I do think this is the reasonable way > of doing it though. > > Note that I'm not advocating removing the start_radar_detection() or its > chandef from the *cfg80211* API. That is clearly needed. But in mac80211 > it seems "set this chandef with radar detection" is a better API? > OK, so we keep the cfg80211 start_radar_detection() and replace the mac80211 part with respective information that radar is required. That could work, I'll look into this. Thanks, Simon
Attachment:
signature.asc
Description: Digital signature