Re: mac80211 flush callback

[Johannes Berg <johannes@sipsolutions.net>]

On Wed, 2012-09-05 at 18:27 +0200, Arend van Spriel wrote:
> On 09/05/2012 03:53 PM, Johannes Berg wrote:
> > Hi Arend,
> >
> >> I ma currently looking into a long standing issue with flush callback in
> >> brcmsmac. After some debugging I found out that mac80211 keeps pushing
> >> packets to brcmsmac during the flush. Is that correct? Should brcmsmac
> >> (or any other driver) stop the mac80211 queues during the flush? My
> >> assumption was that mac80211 would not do transmits during the flush,
> >> but it probably comes from another worker thread.
> >
> > Hmm, good question, this area isn't quite fully worked out yet I
> > think ... probably better to stop queues yourself for now, although I
> > guess mac80211 should really take care to do it ...
> >
> 
> Thanks, Johannes
> 
> I did look at some other driver and it seems they grab a mutex in the 
> flush that is (probably?) also grabbed in the tx path thus blocking it. 
> For brcmsmac I submitted a patch to explicitly stop the queues in the flush.

I don't think any driver could possibly grab a mutex in the TX path
since it needs to be atomic? :)

johannes