On 2019-03-21 3:57 a.m., Stefan Sperling wrote:

> Thanks! Have you confirmed that you can still trigger it in -current
> without my diff?

I have now tried -current without your diff and I can confirm that it
*does not in fact trigger*. All commentary from here on will be in 
relation to unpatched -current

> Does the speed settle eventually? Note that our 11n and 11g modes
> use different Tx scaling algorithms. 11g in particular scales up
> slowly and is sensitive to unrelated traffic on the same channel.
> In 11n mode Tx speed jumps around a lot but should be equal or
> higher than 11g on average.

The speed does not settle, it remains erratic even with quite lengthy
tcpbench testing. I run this exact same hardware as an AP under linux
and have no issues, and I can confirm there is no rouge interference 
either from other APs, or anything else around me with almost total 
certainty.

> Can you show me frames the AP is receiving when the problem occurs?
> This will give you an AP-side packet capture which you could mail to
> me (please compress it with gzip):
> 
>   tcpdump -n -i athn0 -y IEEE802_11_RADIO -s 4096 -w /tmp/athn.pcap
> 

File attached in a separate email to you personally as athn_n.pcap.gz
and athn_g.pcap.gz. Both experience ping and speed issues, but only 
802.11n mode seems to have random problems with auth.

> That might tell us why your client can't associate 50% of the time.
> 
> Do you happen to have an iwn(4) device? If so I'd like to see a packet
> capture in monitor mode. 

Ditto with iwn_g.gz and iwn_n.gz from a third, unassociated system in
monitor mode.

Sorry for the large file sizes, I wanted to capture a decent chunk of the 
behavior, especially the failures to associate. (the athn captures suffered
some dropped packets due to not being able to write fast enough due to a shitty
usb device, can redo tests with an alternate setup if necessary stuff wasn't 
captured)

Thanks again for the effort, I know hardware problems are involve a lot of 
hair pulling.

Reply via email to