linux-backports-modules-jaunty: b43 and ssb regression

[Nick Steeves]

Binary package hint: linux-backports-modules

Background:
I have a PCI BCM4318 which worked flawlessly with Hardy, with the b43 module. I didn't test it when upgrading through Intrepid to Jaunty, so the following might also affect Intrepid. WIth Jaunty it has the forever-repeating:
b43-phy0 ERROR: PHY transmission error
during which the connection becomes unresponsive, or is dropped

The Bug:
Strangely, linux-backports-modules-jaunty-2.6.28-11 and *-2.6.28-13 are even worse. No IP-level link at all. No ping. Route times out like there is a break in the physical layer... Also, very, very strangely, the backports-modules version of ssb and/or b43 now load the pcmcia subsystem -- when it's absolutely not needed for my NIC. This run-of-the-mill Linksys 802.11g PCI NIC isn't one of those PCMCIA/Cardbus-to-PCI adapters.

lspci -v -v reports:

02:01.0 Network controller: Broadcom Corporation BCM4318 [AirForce One 54g] 802.11g Wireless LAN Controller (rev 02)
        Subsystem: Linksys Device 0042
        Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR+ FastB2B- DisINTx-
        Status: Cap- 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
        Latency: 32
        Interrupt: pin A routed to IRQ 22
        Region 0: Memory at f9800000 (32-bit, non-prefetchable) [size=8K]
        Kernel driver in use: b43-pci-bridge
        Kernel modules: ssb

This is wrong...I'm almost certain that it used to say that the b43 kernel driver or module was in use, in addition to the ssb. Additionally, that "b43-pci-bridge" sounds a lot like the a module for the type of card I don't have.

Did the PCI IDs get scrambled somewhere? Please advise which logs you need me to post.

Thanks!
Nick

[Nick Steeves (nick-0)]

Wow was I off with the "b43-pci-bridge" idea! At any rate, I decided to roll a custom, very lean 2.6.30.3 vanilla kernel to use as a baseline (and so I might have something that works if all else fails). b43 works marvelously. I'll post my logs. Next I'll try 2.6.28.10 vanilla and post those. If you would prefer for me to test a different 2.6.28.xy please mention which revision.qq

Then I'll post logs for linux-image-2.6.28-13-generic with linux-backports-modules installed.
And then without. All comments will contain:

1. `uname -a`
2. `dmesg > dmesg.log`
3. `lspci -vvnn > lspci-vvnn.log` (as root)

'hopefully the logs from all of these will encircle and help us to isolate where exactly the problem is. Please advise me which debugging opions I ought to enable. A few backtraces would probably make it much easier, but I don't know anything about kernel-specific debugging...

[Nick Steeves (nick-0)]

1. Linux DigitalMercury.dynalias.net 2.6.30.3-digitalmercury #1 Tue Jul 28 11:58:06 MDT 2009 i686 GNU/Linux
2.

[Nick Steeves (nick-0)]

3. lspci-vnn.log

[Nick Steeves (nick-0)]

P.S. I just remembered that this bug was originally simply going to be to show the differences between 2.6.8.13 and 2.6.8.13-with-backports. Sorry!

bugs 405712, 297959, and 263106 all look like b43 kernel module problems (as opposed to firmware problems)
I'm not sure if they're simply duplicates, or some other type of relation...

[Nick Steeves (nick-0)]

...that's odd... The linux-backports-modules-jaunty b43 driver is working!!! And better than linux-image 2.6.28-13, despite loading that pcmcia, and pcmcia_core -- are we using the pcmcia module as a shim between the updated kernel interface that the updated ssb needs, and the old interface the kernel provides?

But why does it take over 30 seconds to log in with SSH? It should take less than one. This is so slow that nautilus times out when trying to connect with sftp. I tried disabling ipv6 with sysctl for linux-image-2.28-13. I did disable IPv6 in my kernel...

While this is not a regression relative to Jaunty, it's still a significant regression relative to Hardy. More info here: [263106]

[Nick Steeves (nick-0)]

vanilla linux-2.6.28.10, with my minimal baseline config has the
b43-phy0 ERROR: PHY transmission error
and the 30 second lag to log into ssh...and I suspect the connection will soon die

but it has the normal lsmod | grep b43 output
b43 99996 0
mac80211 111884 1 b43
led_class 3844 1 b43
ssb 27396 1 b43

Proposed resolution: backport b43 as close to linux-2.6.30.3 as possible

[Nick Steeves (nick-0)]

...and the connection just died. I didn't let the backports system run long enough to see if it would, but the vanilla linux-2.6.28.10 b43 driver crashed my wireless router. As the backports one still has 30 second TCP delays, it's also useless to me.

[Nick Steeves (nick-0)]

...need to test linux-image-2.6.28-14...which should make this bug obsolete ;-)

[Nick Steeves (nick-0)]

linux-image-2.6.28-14 behaves like a branch of linux-image-2.6.28-13. It does not work, and it fails identically to -13, rather than like the backports-modules for -13

uname -a
Linux DigitalMercury.dynalias.net 2.6.28-14-generic #47-Ubuntu SMP Sat Jul 25 00:28:35 UTC 2009 i686 GNU/Linux

dmesg log attached

[B-ZaR (teemu-erkkola)]

I am having this problem as well. Ubuntu Jaunty, 2.6.28-14-generic, no backports, BCM4318. After installing the firmware, wlan seems to work fine except it's slow. Very, very slow. I don't have the transmission errors in dmesg sten is getting.

dmesg log attached

[Nick Steeves (nick-0)]

B-ZaR, could you please attach the log from the following command?:
sudo bash -c lspci -vvnn > lspci-vvnn.log

[B-ZaR (teemu-erkkola)]

Attachment as requested

[Nick Steeves (nick-0)]

Thanks for the log B-ZaR. I just noticed that I forgot a pair of quotes in that command though... I should have typed:
sudo bash -c "lspci -vvnn" > lspci-vvnn.log

the -vvnn options give much more detailed information; root access is needed to access these details

[B-ZaR (teemu-erkkola)]

Heh, should've noticed the missing quotes. :)

[Nick Steeves (nick-0)]

I'm willing to consider this bug closed, because of improvements to the mainline Linux kernel.

The latest debs of a 2.6.30-series kernel can be found here:

http://kernel.ubuntu.com/~kernel-ppa/mainline/v2.6.30.8/

You'll need the linux-image i386, linux-headers i386, and linux-headers all. What is more, my NIC works great with the Karmic beta. I'm un-subscribing from this bug now.

Best regards,
Nick

[Andy Whitcroft (apw)]

Shove this to the appropriate package.

[Jeremy Foshee (jeremyfoshee)]

As this has most likely been pulled via either rebase or stable updates, I am closing this bug as released. If you find this in error, please reopen and state your case.

Thanks!

~JFo