Timing out external scans?

Ben Greear greearb at candelatech.com
Wed Mar 5 17:03:09 EST 2014


On 03/05/2014 12:46 PM, Jouni Malinen wrote:
> On Wed, Mar 05, 2014 at 11:57:04AM -0800, Ben Greear wrote:
>> I saw a case testing ath10k where it appears the stations would not scan
>> because they thought there were external scans pending.
> 
> It would be interesting to see a wpa_supplicant debug log showing that
> if the state does not disappear automatically.

I watched it happen, but didn't save it..will post it if I see it again.

>> Are those external scans supposed to time out after some time?
> 
> There is no additional timeout for this within wpa_supplicant, but
> cfg80211 is supposed to indicate scan completion (either successful or
> aborted). If that is missing for some reason and this cannot be fixed in
> the kernel side, I guess we may have to consider adding a timeout for
> that within wpa_supplicant as well. The only way I could think of this
> happening today in a known possible way is if you run out of memory and
> the nl80211 event indicating scan completion is dropped. All other cases
> would likely indicate a bug of some sort somewhere (driver, cfg80211, or
> wpa_supplicant).

Ath10k and it's firmware and hardware is definitely buggy.  I have been
reloading drivers, having firmware crash, etc.  But, it also often
recovers...so it would be nice to have supplicant be able to deal
with it as well.

Very unlikely I actually ran out of memory.

>> Once I restarted supplicant the stations immediately scanned and
>> came up fine.
> 
> Please note that this is likely setting the netdev down which is
> clearing a lot of cfg80211 state as well. Without seeing a verbose debug
> log, it is difficult to say where the information about scan completion
> was lost.

Doing ifdown/up and telling wpa_cli to REASSOCIATE did not fix problems,
only restarting supplicant, so it wasn't just clearing some OS state
that fixed things.

I'll go poke in supplicant and try to put in some logic to clean up stale
waiting-for-scan objects...at least in my environment I think this
will be required.

Thanks,
Ben

-- 
Ben Greear <greearb at candelatech.com>
Candela Technologies Inc  http://www.candelatech.com



More information about the HostAP mailing list