Re: AP: ath5k + hostapd occasionally sulks

[Philip Prindeville <philipp_subx@redfish-solutions.com>]

Bob Copeland wrote:
> On Tue, Sep 8, 2009 at 1:53 PM, Philip Prindeville
> <philipp_subx@redfish-solutions.com> wrote:
>   
>> I pulled compat-wireless from GIT last night (or about 1:30am mountain,
>> really) and rebuilt a 2.6.27.29 kernel.
>>
>> I'm seeing a lot of:
>>
>> Sep  8 11:44:09 pbx user.err kernel: ath5k phy0: no further txbuf available, dropping packet
>>
>>
>> one every 10 seconds, in fact.  This is with an Engenius EMP-8062+ card:
>>     
>
> Ok, the timing information is useful.  This is probably something (beacon
> sending?) racing with the periodic calibration, which temporarily stops
> all of the tx traffic and frees the tx buffers, then starts it all up
> again.  In short, apart from the logging this shouldn't cause any
> problems, but we should probably disable the beacon tasklet during this
> time.
>   

Alas it is causing problems.  I have a Windows 7 client with an Atheros
card (I forget which... it's the mini-PCIe card that comes with Zotac
ION mini-itx motherboards).

I either can't associate, or associate but don't get a DHCP address or
don't pass traffic... I forget which.

I can do more testing tomorrow...
> If this only appeared all of a sudden in recent compat snapshots, it
> would be useful to know the last one that worked normally.
>   

I could walk it backwards, I suppose...  2009-08-23 was definitely
working with an 9K board.

I've not tried it with a 5K board (I'm not at this location very often).


>> I'll probably have to reboot regularly, since this is on an embedded box
>> with limited CF filesystem, and I can't overflow my /var partition...
>>     
>
> Ouch.  For now, just take it out or demote it to debug.
>
> As for the original problem, I don't know offhand why a large download
> would trigger a cascade of these errors.  The best way to track it down
> is to try to come up with a case that reproduces it and sprinkle printks
> throughout the driver, especially when we free and allocate the tx
> buffers.
>
>