[rt2x00-users] rt2800pci works nearly fine with 2.6.39rc7 kernel and compat-wireless-2.6.39-rc6-1 - problems during rekeying of PTK

[Helmut Schaa]

Am Dienstag, 31. Mai 2011 schrieb Johannes Stezenbach:
> On Tue, May 31, 2011 at 02:02:10PM +0200, Helmut Schaa wrote:
> > Am Dienstag, 31. Mai 2011 schrieb Johannes Stezenbach:
> > > 
> > > I think hostapd might be waiting for TX_STATUS event, it will trigger
> > > ieee802_1x_tx_status() in hostapd via NL80211_CMD_ACTION_TX_STATUS.
> > 
> > Yeah, but hostapd should output a warning if it experiences a tx status
> > timeout AFAIK. Don't have an example here right now ...
> 
> just above the queue status you quoted from Andreas' log was this:
> 
> May 30 19:01:34 pc hostapd: wlan1: STA 48:5d:60:11:22:33 WPA: sending
> 1/2 msg of Group Key Handshake
> May 30 19:01:35 pc hostapd: wlan1: STA 48:5d:60:11:22:33 WPA: EAPOL-Key
> timeout
> May 30 19:01:35 pc hostapd: wlan1: STA 48:5d:60:11:22:33 WPA: WPA_PTK:
> sm->Disconnect

Hmm, good point. This might be the case, but that doesn't necessarily
mean a tx status was lost. It could also mean the frame was just not
transmitted in time or not answered by the STA in time.

> But I guess I got confused by this thread and there is more
> than one issue. 

Yep, there are at least two different issues (+the one Gertjan already fixed):
1) PTK rekeying fails under high load
2) Sometimes the tx queue seems to get stuck

> However, what does the driver do when
> most TX_STATUS are dropped, but a few are still returned?

As long as the EAPOL frame gets answered by the STA it shouldn't
break rekeying.

> Maybe this could explain why EAPOL fails while data frames
> can still be exchanged? If debug were enabled the driver would
> log the missing TX status, but the queue would not get completely
> stuck, right?

No, rt2800pci currently doesn't use the watchdog and hence won't detect
lost tx status reports. I once started to write a watchdog but never finished
it since I could't make it fail any reports anymore.

But without further investigation I tend to believe that no tx status
is lost but the frames are just not transmitted in time (for whatever reason),
but that is only a gut feeling.

Let's first see if Andreas can reproduce the stuck tx issue with HT20.

Helmut