System software teleconference minutes - 2007-04-03

Mitch Bradley wmb at
Tue Apr 3 22:26:10 EDT 2007

System Software Teleconference, 20070403

Present: Mitch, Jim, Jordan, Chris Ball, Andres,
Andrey, David (Quanta), Marcelo, J5, Ronak,

New Action Items:

AI: Dan - make a wiki page describing manual/fallback wireless configuration

AI: Wad - wiki-link the doc for the libertas driver ioctl

AI: jg - find out where to plug in the font alias for Adobe.

AI: wmb - study what is necessary to bring back touchpad after resume

Old action items:

AI: dcbw and marcopg to investigate mesh/avahi not working.
      done, but we're seeing crashing reports

   Done - Avahi was starting up too early and seeing the hostname as 
   The patch might not have made it into 368.

AI: Richard to work with Ronak  & Michailis to get him the wireless
wakeup info he needs.



April 15 is the scheduled build B3 proto build date.  Would like OLPC people
to appear on April 18.  Mitch and Richard will be there (Richard via
Singapore), Dave W. will arrive the following week (Apr 23).

NAND FLASH parts have arrived in Cambridge and have been sent out to
various people (e.g. Mitch).

Several last-minute firmware spins were done for the B2-2 build, 
EC code fixes for LiFePO4 battery issues.

We have some problems with WiFi access points.  Under investigation.

Also some problems with strange X crashes.

No major known issues with Libertas firmware.

When people see association problems, it would be useful to try running
iwconfig manually.

There are some issues surrounding the use of wireless devices when wired
devices are present.

Dan: The problems talking with school server were caused by network manager.
The replies were being rejected because of a NM bug.  There is still
a problem with dhclient.

jg: Is going to a trial school next week in Porto Allegre.

We will start the cutover to C-series firmware tomorrow.  Will use the 
tree for configuration.

* Blocker bugs for trial-1

* JFFS2 problem status

  David found the problem with JFFS2.  It is memory exhaustion because
  a log file that was only being created on Michalis's machines was
  generating zillions of tiny nodes.

* Sugar crashes/hangs

  It seems to happen every 15 to 20 reboots.  It can be triggered by 
  Sugar, restarting, then mouse-over.  It seems to be a double-free in 
  May need Behdad or Keith to look at it.  First try later release - 
242.  There
  is a segfault bug that is reported to have been fixed in mmap shared 
fontconfig data.

* Mesh status - how to better test with people, rather than artificial 
test jigs?

  Dan has some ideas for testing meshing.  Something about blinding 
table and MPPs.

  Michaelis has discovered that the routing prefers hops, picking routes 
that have
  more hops than necessary.  He has a test scheme that involves having a 
  connection too.

  We haven't tested what happens at the user level when roaming.

  Cozybit is working on a tool to display route maps.  It uses the 
blinding table
  and the inverse blinding table.

* Priorities
  * Test-1 requirements
    critical bugs.

        server - wad
        Backup needed for laptops to servers urgently

    We need to get a script.  Now that we are getting a mesh, it is 
    to get going.  It may require DHCP.

    Adobe Flash crash seems to be a font problem.

    Marco and co. are looking into why the browser has been crashing.

    HW bug: if you see power off right after the kernel is loaded,
    add the serial # to trac #1256 .

* Suspend/resume

  Resume currently takes about 250 mS (went from 900 to 200+ simply by
  turning off serial output from kernel logging).  Everything is resuming
  except SD.

  The touchpad is not working on resume, on a system that has been
  modified to retain the power.  It just started happening with a
  recent kernel patch that stops bouncing the keyboard.  jg points out
  that a recalibrate might be necessary - 4-finger salute might help.

Plan for lxfb driver: Jordan will be coding "blindly" before and during
his office move.  Some testing may have to be done by others.

More information about the Devel mailing list