WRT54g / b43 / mac802.11 BREAKTHROUGH

[Hauke Mehrtens]

On 08/22/2012 07:49 PM, Rafał Miłecki wrote:
> 2012/8/22 Hauke Mehrtens <hauke at hauke-m.de>:
>> On 07/18/2012 01:56 PM, Bastian Bittorf wrote:
>>> hi devs!
>>>
>>> yesterday we had a breaktrough debugging b43 in our hackspace maschinenraum/m18[1,2]
>>> at weimar.freifunk.net[3,4] - since a long time our darling wrt54g suffers from a
>>> hanging wifi and bad performance[5], but the workaround is easy: now it's up to
>>> you to fix the rootcause.
>>>
>>> our testsetup, where we could _reproduce_ reliably a stopping TX is like this:
>>>
>>> laptop ---lan--- "A"-wrt54g/adhoc ~~~   air  ~~~ "B"-anyrouter/adhoc
>>>
>>> now make a testdownload with the laptop from B.
>>> wireless will stop working after ~10 seconds.
>>>
>>> wifi up will reanimate and our freifunk-cron.minutely-check
>>> will do this automagically. (read further, this is not the solution)
>>>
>>> we tried to limit the rateset to e.g. lower rates, but this does NOT change
>>> the behaviour. what works is: define a rateset on BOTH router which makes
>>> it impossible to change the band, e.g.:
>>>
>>> iw dev $WIFIDEV set bitrates legacy-2.4 1 2 5.5 11
>>> OR
>>> iw dev $WIFIDEV set bitrates legacy-2.4 6 9 12 18 24 36 48 54
>>>
>>> now we had a great performance, 10 Gigabytes of wireless transfer,
>>> no stopping TX anymore and an empty box of beer. three things to do now:
>>>
>>> 1) why does a band change (can be seen through minstrel) is a problem?
>>>
>>> 2) we have to make in config-option to force a band, or a rateset:
>>>       e.g. uci set wireless.radio0.hwmode=11g
>>>       e.g. uci set wireless.radio0.bitrates='6 9 12 18 24 36 48 54'
>>>
>>> 3) spread to word:
>>>    there is a great frustration in the community about b43,
>>>    but the old drivers _have_ to die, it's about time - really.
>>>
>>> thanks for your work,
>>> bye storchi, andi, bastian + m18 crew
>>>
>>> [1] http://blog.maschinenraum.tk/
>>> [2] http://blog.maschinenraum.tk/2012/07/15/bitcoin-vending-machine-exchange-euro-coins-for-bitcoin-wallets/
>>> [3] http://wireless.subsignal.org
>>> [4] http://wireless.subsignal.org/index.php?title=Main_Page_en
>>> [5] https://dev.openwrt.org/ticket/7552
>>
>> I tried this on some of my devices (BCM4318 G-PHY and BCM4322 N-PHY) and
>> it is easily reproducible on all tested devices when restricting the
>> rates to 11 and 12 MBit/s.
>>
>> I have the Broadcom device working as an Access point (on a MIPS SoC)
>> and a Laptop with an Intel Wifi device is connected to it. I generated
>> the traffic with iperf. If the Access Points sends the traffic the
>> problem occurs, if it is just receiving there is not problem.
>>
>> After the problem occurred b43_generate_txhdr is called rarely ( every
>> ~10 seconds) and I am not able to stay connected or connect to the
>> network any more. I am not familiar with the internal flow in b43 or
>> mac80211 could someone give me a hint where to look.
>>
>> I can not see any special changes between CCK and OFDM rates before it
>> goes down there are many changes without a problem before it goes down.
>>
>> Currently I do not have a Broadcom wifi card running in a x86 device
>> just mips devices could someone try to reproduce the problem on a x86
>> device.
>>
>> I added the b43 mailing list and Rafał.
> 
> Does your kernel include
> 
> commit bad6919469662b7c92bc6353642aaaa777b36bac
> Author: francesco.gringoli at ing.unibs.it <francesco.gringoli at ing.unibs.it>
> Date:   Fri Dec 16 18:34:56 2011 +0100
> 
>     b43: avoid packet losses in the dma worker code.
> 
> ? I have a lot of things on my TODO list, including that. Right now I
> just want to focus on BCM4706 support, which slowly moves into
> mainline.
> 

This patch is included. I used compat-wireless-2012-07-16 plus some more
backport patches for b43.