Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 13 Apr 2014 23:22:26 GMT
From:      Adrian Chadd <adrian@freebsd.org>
To:        freebsd-gnats-submit@FreeBSD.org
Subject:   kern/188576: [ath] traffic hangs in station mode when downgrading from AMPDU TX or reassociating
Message-ID:  <201404132322.s3DNMQmU076513@cgiserv.freebsd.org>
Resent-Message-ID: <201404132330.s3DNU001065782@freefall.freebsd.org>

next in thread | raw e-mail | index | archive | help

>Number:         188576
>Category:       kern
>Synopsis:       [ath] traffic hangs in station mode when downgrading from AMPDU TX or reassociating
>Confidential:   no
>Severity:       non-critical
>Priority:       low
>Responsible:    freebsd-bugs
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Sun Apr 13 23:30:00 UTC 2014
>Closed-Date:
>Last-Modified:
>Originator:     Adrian Chadd
>Release:        HEAD
>Organization:
>Environment:
FreeBSD lucy-11i386 11.0-CURRENT FreeBSD 11.0-CURRENT #1 r263418M: Tue Apr  1 11:33:21 PDT 2014     adrian@lucy-11i386:/usr/home/adrian/work/freebsd/head/obj/usr/home/adrian/work/freebsd/head/src/sys/LUCY_11_i386  i386

>Description:
Whenever an ath(4) 11n station reassociates or downgrades from aggregation to no aggregation, there's a chance that it'll hang and refuse to queue more frames.

The session needs to be fully torn down (eg ifconfig wlanX down) for things to go back to normal.
>How-To-Repeat:

>Fix:
I actually have debugged this a little already.

So the problem seems to be that there's more than one entry point into ath_tx_tid_cleanup(). It's likely a couple of calls into the reassociation path or one into reassociate and one into aggregation teardown. I'll go figure that bit out soon.

But what it leads to is thus:

* the caller causes ath_tx_tid_pause();
* ath_tx_tid_cleanup() is called;
* the first time this happens it sees there's 1 or more frames to cleanup, so it sets tid->cleanup_inprogress;
* the caller then checks if that's set to 1 - if so, it assumes that it should wait until the cleanup is finished;
* otherwise it calls ath_tx_tid_resume().

If tid->cleanup_inprogress is set to 1 then the normal TX completion path will eventually call ath_tx_comp_cleanup_unaggr() or ath_tx_comp_cleanup_aggr() which will clear the flag and resume the TID.

If a second path through ath_tx_tid_cleanup() occurs, then:

* the caller pauses;
* ath_tx_tid_cleanup() is called;
* tid->cleanup_inprogress is set to 1, but there's no code to check whether this call actually set it or not - so it doesn't call ath_tx_tid_resume().

So once the frames complete and ath_tx_tid_resume() is called, there's still a pending paused reference and thus traffic never continues flowing.


>Release-Note:
>Audit-Trail:
>Unformatted:



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201404132322.s3DNMQmU076513>