Comment 33 for bug 965043

The Kernel Bisect is finally completed.

I found the problem was caused by this commit:

commit f0425beda4d404a6e751439b562100b902ba9c98
Author: Felix Fietkau <email address hidden>
Date: Sun Aug 28 21:11:01 2011 +0200

    mac80211: retry sending failed BAR frames later instead of tearing down aggr

    Unfortunately failed BAR tx attempts happen more frequently than I
    expected, and the resulting aggregation teardowns cause performance
    issues, as the aggregation session does not always get re-established
    properly.
    Instead of tearing down the entire aggr session, we can simply store the
    SSN of the last failed BAR tx attempt, wait for the first successful
    tx status event, and then send another BAR with the same SSN.

    Signed-off-by: Felix Fietkau <email address hidden>
    Cc: Helmut Schaa <email address hidden>
    Signed-off-by: John W. Linville <email address hidden>

Upon further investigation, I have found some discussion on the same issue:

http://marc.info/?l=linux-wireless&m=132534063601847&w=2

And a possible patch to the rt2x00 driver to fix the issue:

http://permalink.gmane.org/gmane.linux.drivers.rt2x00.user/569

I will apply the patch to my rt2x00 drivers later today and report back my findings...