Degradation on b43 14e4:4315 ?

Thu Jun 16 18:08:12 EDT 2011

W dniu 16 czerwca 2011 23:23 użytkownik Oncaphillis
<oncaphillis at snafu.de> napisał:
> On 06/16/2011 11:22 AM, Rafał Miłecki wrote:
>>
>> Hey,
>>
>> 2011/6/16 Oncaphillis<oncaphillis at snafu.de>:
>>>
>>>  I'm using my b43 14e4:4315 on my acer notebook for quite a while now.
>>> I've started to use a wireless git-kernel and helped a bit in debugging
>>> initial problems with this device. Since the early Fedora 14 it seemed to
>>> work out of the box, but now when Fedora 14 moved to 2.6.35.12+ kernels
>>> I loose connections on heavy transfer.
>>
>> Do you get something interesting from "dmesg | grep b43"?
>>
>
> So I got a vanilla kernel 2.6.35.13 with the .config stolen from the
> Fedora 14 kernel except that I disabled DNA. On startup dmesg tells
> me:
>
> <snip>
> [   11.038578] b43-pci-bridge 0000:01:00.0: PCI INT A -> GSI 16 (level, low)
> -> IRQ 16
> [   11.038643] b43-pci-bridge 0000:01:00.0: setting latency timer to 64
> [   13.725758] b43-phy0: Broadcom 4312 WLAN found (core revision 15)
> [   13.740378] b43-phy0 debug: Found PHY: Analog 6, Type 5, Revision 1
> [   13.740448] b43-phy0 debug: Found Radio: Manuf 0x17F, Version 0x2062,
> Revision 2
> [   14.024423] Registered led device: b43-phy0::tx
> [   14.024523] Registered led device: b43-phy0::rx
> [   14.024613] Registered led device: b43-phy0::radio
> [   25.399187] b43-phy0: Loading firmware version 478.104 (2008-07-01
> 00:50:23)
> [   25.401793] b43-phy0 debug: b2062: Using crystal tab entry 19200 kHz.
> [   25.479462] b43-phy0 debug: Chip initialized
> [   25.479631] b43-phy0 debug: PIO initialized
> [   25.479660] b43-phy0 debug: QoS disabled
> [   25.487963] b43-phy0 debug: Wireless interface started
> [   25.492355] b43-phy0 debug: Adding Interface type 2
> </snip>

I guess you chose B43_FORCE_PIO == "Force usage of PIO instead of DMA"?

I suggest you don't select this, but use module parameter instead.
Don't mess with B43_FORCE_PIO but use:
rmmod b43; modprobe b43 pio=0
rmmod b43; modprobe b43 pio=1

> And the connection stays stable. But If I do some heavy file transfer
> I loose connection and dmesg tells me:
>
> <snip>
> [ 7531.098190] cfg80211: Calling CRDA to update world regulatory domain
> [ 7531.099543] cfg80211: Calling CRDA for country: DE
> [ 7531.673634] cfg80211: Regulatory domain changed to country: DE
> [ 7531.673642]     (start_freq - end_freq @ bandwidth), (max_antenna_gain,
> max_eirp)
> [ 7531.673650]     (2400000 KHz - 2483500 KHz @ 40000 KHz), (N/A, 2000 mBm)
> [ 7531.673656]     (5150000 KHz - 5250000 KHz @ 40000 KHz), (N/A, 2000 mBm)
> [ 7531.673663]     (5250000 KHz - 5350000 KHz @ 40000 KHz), (N/A, 2000 mBm)
> [ 7531.673669]     (5470000 KHz - 5725000 KHz @ 40000 KHz), (N/A, 2698 mBm)
> [ 7532.279452] wlan0: authenticate with 00:25:9c:d0:04:fb (try 1)
> [ 7532.479109] wlan0: authenticate with 00:25:9c:d0:04:fb (try 2)
> [ 7532.679347] wlan0: authenticate with 00:25:9c:d0:04:fb (try 3)
> [ 7532.879169] wlan0: authentication with 00:25:9c:d0:04:fb timed out
> [ 7548.738780] wlan0: direct probe to 00:25:9c:d0:04:fb (try 1)
> [ 7548.939078] wlan0: direct probe to 00:25:9c:d0:04:fb (try 2)
> [ 7549.138181] wlan0: direct probe to 00:25:9c:d0:04:fb (try 3)
> [ 7549.338067] wlan0: direct probe to 00:25:9c:d0:04:fb timed out
> </snip>
>
> No messages from the b43 module though. In order to reanimate wlan
> I just have to kill wpa_supplicant and do dhclient startup. No
> removal and reinsert of module necessary. May be I bark up the wrong
> tree ? And it's wpa_supplicant that does it all wrong ?

Please, try using DMA and provide us dmesg | grep b43 after seeing
problems. I believe DMA has more advanced debugging implemented.

>>> Where there developments during the 2.6.35.x developments which could
>>> explain such behaviour ?
>>
>> Do you mean 2.6.35 was working fine and 2.6.35.12 already contains
>> regression? Were you using clean 2.6.35 earlier with success?
>
> I started up with the wireless git-repo and somewhere since the early
> Fedora 14 it worked for me with the fedora standard kernel. Throughput could
> have been better, but I never got such disconnections ever since. Actually I
> don't know if Fedora 14 started up with kernel 2.6.35 or even earlier.

http://distrowatch.com/table.php?distribution=fedora says kernel
2.6.35.6 is default in Fedora. I didn't check for kernel history log
yet, but I don't believe anything could change between 2.6.35.6 and
2.6.35.12 that is important for b43.

After testing "DMA" and providing dmesg | grep b43, I'd like you to
try switching back to older firmware.

How to install older firmware:
1) Move newer firmware to other dir:
mv /lib/firmware/b43 /lib/firmware/b43-478.104

2) Download older firmware and install it:
wget http://mirror2.openwrt.org/sources/broadcom-wl-4.150.10.5.tar.bz2
tar xjf broadcom-wl-4.150.10.5.tar.bz2
sudo b43-fwcutter -w /lib/firmware broadcom-wl-4.150.10.5/driver/wl_apsta_mimo.o

-- 
Rafał