Re: [TIP] BUG kmalloc-4096: Poison overwritten (ath5k_rx_skb_alloc)

[Bob Copeland <me@bobcopeland.com>]

On Fri, May 22, 2009 at 10:39:31AM +0100, Sitsofe Wheeler wrote:
> The poison message has not reappeared but this morning there was a
> sudden break in the connection. Only toggling the wifi via rfkill
> brought things back to life. dmesg below:
> 
> [   71.445152] wlan0: associated
> [ 5771.345217] ath5k phy0: noise floor calibration timeout (2412MHz)
> [ 5774.632085] wlan0: no probe response from AP 00:01:38:d6:0b:4f - disassociating

Card got into a bad state here somehow, 

> [ 5774.869703] ath5k phy0: failed to wakeup the MAC Chip
> [ 5774.869717] ath5k phy0: can't reset hardware (-5)

-EIO, a really bad state (we're doing a lot of resets, not sure
offhand why -- maybe just the calibration timer since it doesn't
look like the channel is changing as would happen in a scan).  

When we see this either card isn't getting any power (e.g. due
to rfkill, but I think that happened later here) or card just
isn't responding.  Seems we should give up after a while rather
than spamming the log forever.

> [ 5834.122061] ath5k 0000:01:00.0: PCI INT A disabled

I take it this is when you hit rfkill, which also did an rmmod and
probably disconnected it from the bus?

> [ 5836.901310] pci 0000:01:00.0: reg 10 64bit mmio: [0x000000-0x00ffff]

and here's where you turned it back on...

> Another weird thing that I've noticed with 2.6.30 kernels (although it
> might be happening with older ones) is that periodically (maybe once a
> day) the packet loss between the computer and the access point will
> become very high and will stay that way until the computer is rebooted.
> I do not know if toggling via rfkill will also solve that problem.

It sounds like a similar case.  There's a patch on the way (I don't
think has hit 2.6.30 yet) that fixes a regression that can affect
calibration in 802.11g mode.

-- 
Bob Copeland %% www.bobcopeland.com