OLPC System Software Telecon Minutes
2006-09-12

Jim, Mitch, Ron, Tom, Ivan, Ray, Richard, Jordan, Chris Ball, Andres Solomon, Marcello


Introducing new team member Andres Solomon, Debian Package Maintainer, kernel hacker, musician,
rock climber.  Welcome Andres!

Several of us have functioning displays.  4 of 7 were damaged in shipment.

The display quality is good.  See picture on web site.

RAM timing problem: Infineon shipped some of the wrong part number to Quanta.

Ron:  NRV uncompress is working, starting on LZMA.  Need to get the clock rate fix in (use 366 timing).
We are going with CL2 because the design is qualified for that timing assuming that the right parts are used.
Richard's warmboot patch is committed.

Tom: Making a list of power than can be controlled.  Going to edit the wiki page.

Jim: we have two dead boards.  How do we do failure analysis.  So does Tom and Marcello.

Send dead boards to Mitch Bradley, 615 Olinda Road, Makawao, HI 96768
Mitch needs an FS2 thing.  Tom knows about FS2.  Need some stuff in /etc/hotplug/usb

Chris and Ivan:

Tinderbox:  Python scripts to audit software builds and installation, working toward an automated
test suite for the software.  Also want to measure boot times.  Serial port output time could influence
the results.

Tinderbox has been used very effectively in the Mozilla project.

Ray:
Infineon status: shipped 150 boards out of 500.  We don't have the serial numbers.

Ivan has run Memtest for 56 hours on an Infineon board, but at CL2.5 .

Maybe we could overclock to quickly sniff out Infineon parts.

Discussion of techniques to auto-detect Infineon parts.

Richard to prepare two LB images, identical but for CL.

Ivan wants to clone the RH build process at the OLPC offices.  David Z. is the person to talk to.
AI: Ivan and Marcello to bug David Z. document the build process so Ivan can win.

Ray: How to write serial numbers into SPI ROM?  Wide-ranging discussion of exactly when to write-protect
the SPI in the manufacturing process.  Conclusion: Ray and Mitch own the issue, Ivan to review the solution.
Data points:
* Serialization occurs at a manufacturing phase distinct from and later than the burnin phase.
* Quanta wants to reserve a 2K chunk for serialization info
* Richard: Different parts behave differently with respect to their write-protect defaults