authentication timeouts with ath(4) in hostap mode

[Petar Bogdanovic]

On Fri, May 02, 2008 at 05:25:09PM -0700, Sam Leffler wrote:
> Petar Bogdanovic wrote:
>> Hi,
>>
>> I'm using an alix2c0 board with two winstron CM9 ath(4)-cards and
>> FreeBSD 7:
>>
>> 	ifconfig ath0 (...) mediaopt hostap mode 11a channel 36 ssid sn.a -bgscan
>> 	ifconfig ath1 (...) mediaopt hostap mode 11g channel 11 ssid sn.g -bgscan
>>
>>
>> When I try to raise the traffic (i.e. dd | ssh AP dd) my Linux
>> wpa_supplicant drops the connection and has to reassociate. This however
>> does not work immediately; The supplicant fails a few times before
>> reconnecting:
>>
>> 	<2>CTRL-EVENT-CONNECTED - Connection to 00:0b:0b:06:0d:09 completed (reauth) [id=0 id_str=]
>> 	<2>CTRL-EVENT-DISCONNECTED - Disconnect event - remove keys
>> 	<2>Trying to associate with 00:0b:0b:06:0d:09 (SSID='sn.a' freq=5320 MHz)
>> 	<2>Authentication with 00:0b:0b:06:0d:09 timed out.
>> 	<2>Trying to associate with 00:0b:0b:06:0d:09 (SSID='sn.a' freq=5320 MHz)
>> 	<2>Authentication with 00:0b:0b:06:0d:09 timed out.
>> 	<2>Trying to associate with 00:0b:0b:06:0d:09 (SSID='sn.a' freq=5320 MHz)
>> 	<2>Authentication with 00:0b:0b:06:0d:09 timed out.
>> 	<2>Trying to associate with 00:0b:0b:06:0d:09 (SSID='sn.a' freq=5320 MHz)
>> 	<2>Authentication with 00:0b:0b:06:0d:09 timed out.
>> 	<2>Trying to associate with 00:0b:0b:06:0d:09 (SSID='sn.a' freq=5320 MHz)
>> 	<2>Authentication with 00:0b:0b:06:0d:09 timed out.
>> 	<2>Trying to associate with 00:0b:0b:06:0d:09 (SSID='sn.a' freq=5320 MHz)
>> 	<2>Authentication with 00:0b:0b:06:0d:09 timed out.
>> 	<2>Trying to associate with 00:0b:0b:06:0d:09 (SSID='sn.a' freq=5320 MHz)
>> 	<2>Associated with 00:0b:0b:06:0d:09
>> 	<2>WPA: Key negotiation completed with 00:0b:0b:06:0d:09 [PTK=CCMP GTK=CCMP]
>> 	<2>CTRL-EVENT-CONNECTED - Connection to 00:0b:0b:06:0d:09 completed (reauth) [id=0 id_str=]
>>
>>
>> This happens more on the 11a than on the 11g network. When I'm next to
>> the AP, the timeouts are almost gone but they still happen. (My laptop
>> is just one room away from the AP). Here is the athstats-output of ath0
>> (11a):
>>
>> 	# ./athstats -i ath0
>> 	481546 data frames received
>> 	330669 data frames transmit
>> 	13395 tx frames with an alternate rate
>> 	78558 long on-chip tx retries
>> 	1431 tx failed 'cuz too many retries
>> 	36M current transmit rate
>> 	78 tx management frames
>> 	3 tx frames discarded prior to association
>> 	45 tx frames with no ack marked
>> 	2894 rx failed 'cuz of bad CRC
>> 	2 rx failed 'cuz decryption
>> 	92711 rx failed 'cuz of PHY err
>> 	    92708 OFDM timing
>> 	    3 OFDM restart
>> 	318332 beacons transmitted
>> 	1111 periodic calibrations
>> 	2 rfgain value change
>> 	22 rssi of last ack
>> 	23 avg recv rssi
>> 	-96 rx noise floor
>> 	2530 switched default/rx antenna
>> 	Antenna profile:
>> 	[1] tx   173364 rx   123068
>> 	[2] tx   155874 rx   358671
>
> So the obvious question is whether your system config has enough isolation 
> of the radios for them not to impact each other?  I have no experience with 
> Alix boards but it's not uncommon for there to be power and signal issues 
> when operating multiple radios in an enclosure (and yes, even with the 
> radios on different bands).
>
> You don't indicate what you've done to diagnose this problem.  Have you 
> verified the packets are present in the air?  Have you traced packets 
> and/or phy errors around the time of the problem?  Does turning off one 
> radio give you stable operation?  Have you tried different channels? Have 
> you tried different boards?
>
>
>>
>>
>> All this is well known to me, since I had NetBSD running on this device
>> for months and it suffered the same problems -- it was even worse, the
>> timeouts occured every few minutes. Back then, it seemed that ath had
>> some interrupt problems:
>>
>> 	ath0: device timeout
>>
>> as David Young from NetBSD noticed in his mail some time ago:
>>
>> 	http://mail-index.netbsd.org/tech-net/2007/11/29/0001.html
>>
>>
>> FreeBSD doesn't seem to have this `device timeouts'. I don't see any in
>> /var/log/messages and there are none when I'm connected to the device
>> over a serial port.
>>
>> I'm a bit lost here, but ready to debug if someone knows more.
>
> netbsd's code base is many _years_ out of date wrt freebsd; comparing 
> operation of the two systems is unlikely to be useful.

Just for the record:

After various (client-)tests with Intel 2200BG and 3945ABG chips and one
AR5212 chip on FreeBSD, it seems that only the madwifi client caused
interrupts or more precisly: interrupts after a missed beacon.

My observations correspond with the following madwifi ticket:

	http://madwifi.org/ticket/848

One beacon miss, one interrupt. The FreeBSD ath-driver does _not_ show
any similar behaviour on the same device.


Kind regards,

Petar