Date: Thu, 9 Aug 2012 12:54:34 -0400
From: Marco Fonseca <marco@tampabay.rr.com>
To: linux-wireless@vger.kernel.org
Subject: carl9170: Unintentional transmission in monitor mode?
Message-ID: <20120809165434.GA13409@192.168.1.10> (sfid-20120809_190810_334694_53BADD01)
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Sender: linux-wireless-owner@vger.kernel.org

Hello All,

I've noticed unexpected transmissions from the AR9170 while in monitor mode
and using tcpdump (and sometimes without tcpdump running).  These
transmissions degrade (pretty significantly) performance of other STA/AP on
the channel (as seen by iperf going from 60mbit/s to almost nill).  As best I
can tell these transmissions never coalesces into an actual frame in the air
(at least not one I'm able to pick up with a second different adapter in
monitor mode).

Are there any known reasons/bugs on this?  Is there anyway to completely
silence the transmitter of the AR9170 while still allowing reception?


Some additional information that may give some context:
- This seems to happen over a wide range of releases of compat wireless I've
  tried, I've tested between compat wireless 2.6.39-rc6-1 and 2012-7-03,
  various drivers and firmware (the latest being driver 1.9.4 and firmware
  1.9.5).  I've also seen the problem across multiple AR9170 adapters, as well
  as different linux hosts.

- I've put a few printk's in the carl driver & stubbed out (somewhat blindly)
  some of the tx functions in wlan.c in the firmware to see if I could silence
  or get a clue on whats going on, with not much success.  The problem still
  occurred (No prink's told me the kernel wasn't actively generating any
  pkts).

- Seems like the transmissions only occur when monitoring 802.11n speeds.
  (That is if I manual configure the AP to 54mbit/s, I don't see the
  issue).  Happens in both 2.4 or 5ghz band.

- I was able to also observe this behavior with a single ping.  Only seemed to
  occur with larger ICMP pings (would occur with -s 2000 pings, but not a
  standard (56 byte) ping).

- Along with iperf degradation, I've verified transmissions with a spectrum
  analyzer.


Thank you,

Marco Fonseca