#12140 NORM Not Tri: XO-1.5 WLAN sometimes fails to be power cycled when firmware stops responding

[Zarro Boogs per Child]

#12140: XO-1.5 WLAN sometimes fails to be power cycled when firmware stops
responding
----------------------+-----------------------------------------------------
 Reporter:  rsmith    |                 Owner:  dsaxena                 
     Type:  defect    |                Status:  new                     
 Priority:  normal    |             Milestone:  Not Triaged             
Component:  wireless  |               Version:  Software Build 12.1.0-21
 Keywords:            |           Next_action:  never set               
 Verified:  0         |   Deployment_affected:                          
Blockedby:            |              Blocking:                          
----------------------+-----------------------------------------------------
 I discovered this while continuing my on-going saga of trying to make a
 power test that can do lots of suspend resumes followed by network access.

 Attached is my last version of netsus where I've been trying to figure out
 how to make a script that can DTRT when the network dies.

 I finally got around to running it on some of the 1.5s I have at my
 apartment.  One of the units was running 11.3 and the other 12.1.  I
 noticed that the 11.3 machine would run for hundreds of my loops with no
 problems.  It recovered nicely.  The 12.1 machine however would seldom run
 more than 30 cycles without aborting when my network check decides that
 the network is gone and not coming back.

 It was very confusing for awhile because I would return to find my test
 halted but then when I inspected the machine everything was ok.  If I re-
 ran my test it started working.  What I finally discovered was happening
 was that when my test stops it allows powerd to take back over.  Powerd
 then detects idle and puts the system into suspend.  At that time the wlan
 is broken and not associated to an AP so the power to the wlan card will
 be turned off.  So on resume you have power cycled the card and it comes
 back functional.

 I didn't remember my problems before being specific to an OS release so I
 was thinking the difference between the machines was hardware variations
 but decided to try 12.1 on the machine that was running 11.3 just to be
 sure. Bingo, 28 loops later it stopped.

 This does not seem to duplicate on a 1.75 12.1 system.  That or the
 frequency is very low.  My test machine has run over 300 loops without
 error.

 I've included the dmesg outputs that have my netsus: debug messages in
 them.

 'wlan_fail.log' is the fail from the 1st 12.1 machine.

 '12.1_fail.log' is from the 2nd machine that was known good working under
 11.3.  I stopped my script while it was in the retry phase so I could
 examine the network and found eth0 was indeed missing.

 '12.1_fail_post_pwrcycle.log' is immediately after the 12.1_fail.log but I
 allowed powerd to suspend the machine first.

-- 
Ticket URL: <http://dev.laptop.org/ticket/12140>
One Laptop Per Child <http://laptop.org/>
OLPC bug tracking system