On 16 January 2014 10:31, Johannes Berg <johannes@xxxxxxxxxxxxxxxx> wrote: > On Thu, 2014-01-16 at 10:28 +0100, Michal Kazior wrote: >> On 15 January 2014 14:22, Johannes Berg <johannes@xxxxxxxxxxxxxxxx> wrote: >> > On Wed, 2014-01-15 at 13:04 +0100, Michal Kazior wrote: >> >> This implements a fairly simple multi-interface >> >> CSA. It doesn't support multiple channel >> >> contexts so it doesn't support multi-channel. >> > >> > It seems that this also means that somewhere you need to advertise the >> > multi-switch feature. I guess you need that anyway, but this makes it >> > obvious. >> >> I don't really understand. Do we really need it? You want to advertise >> it just to cfg80211 or all the way up to userspace? > > It seems it would be needed all the way to userspace so it knows whether > it can expect multi-vif CSA to be available? Or would you just want to > try & fail? Wouldn't that potentially impact channel selection since > you'd want to not pick radar channels for multi-vif? Not really sure ... > > Maybe the expectation is that internally there will be no drivers that > don't support it, i.e. if a driver implements the cfg80211 API at all > then it must support multi-vif CSA? Typically if you run more than one AP on a DFS channel and you detect a radar you want to move all APs - so you either succeed with multi-vif CSA or you fallback and re-start them. If you assume a driver supports multi-channel but isn't capable of multi-interface CSA you'll be running into timing issues. After you detect radar you have a limited time to quiesce. You might end up exceeding that limit if you move APs one-by-one. With my patches as they are new hostap will try sending new channel switch command but since it doesn't contain ifindex old kernel will respond with a -EINVAL and userspace has no way of knowing if the command was actually malformed or isn't supported. hostap could still, however, work in a best effort manner and fallback to the old command variant. Michał -- To unsubscribe from this list: send the line "unsubscribe linux-wireless" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
