[OLPC-devel] Minutes from System Software telecon, 2006-08-22

[Mitch Bradley]

OLPC System Software meeting minutes, 2006-08-22

Present: Ray, Jim, Jonathan, Zephaniah, Mitch (scribe), Jordan,
Richard, Stiven, Chris Ball, Marcello, Chris Blizzard, Ivan

Schedule update: touchpad expected at end of month, Zephaniah on hot seat
CaFe ahead of schedule, FPGA bringup planned for Sep 6, 2nd bringup on
other FPGA about Sep 20.  About 2 week delay in Btest build because of
DCON chip change to fix pixel problem.

David's experience/status of distro: unblocked on memory exhausted problem.
Issues: getting the right drivers, integrating X driver, cleanups.

Must get X driver mode right
Get Sugar working.
Don't have the LinuxBIOS and flashrom stuff into Fedora.

Action Item: Marcello to integrate and package the
olpcflash stuff into Fedora.

Tests:
a) Boot on Insyde, both flash key and USB hard drive
b) Boot on LinuxBIOS, both flash key and USB hard drive

NAND FLASH support is not a hard criterion.
- Booting from internal NAND FLASH doesn't work because
  the NAND and JFFS2 drivers are not in the payload kernel.

Marcello will send the NAND Flash patch to Richard Smith.
Richard can't commit to ultra-quick response because of job commitments.
Richard will try to test it tonight.  The code that is in
Fedora is good enough.  Marcello will send Richard a RedHat
image to write into NAND.

Ivan's login problem: nondeterministic, Ivan hopes to make
progress tonight.

Chris: new image targeted for 10 tomorrow morning.

RedHat build system is being difficult.

Action Item:  Marcello and David Z. will update the instructions for
how to reflash the SPI.

Action Item: Jim will make the announcement

Jordan will make LB print something on the screen - the build
date and time.  We need to make a symbolic tag for the Commit ID
of the LinuxBIOS version.

Mitch is release czar for LinuxBIOS.  Richard will supply bits.

Chris Blizzard is release czar for Fedora bits.

Jim is the gatekeeper.

Manufacturing test:

Text only user memory, or test all memory?

Consensus that we need something like memtest86.

If the burn-in test runs for 4 hour, they would
devote, say, 1 hour for memory test.

3 places to put memory test:
NAND FLASH
LinuxBIOS payload
LB init code

Distinction between design verification tests and unit tests.

Burn-in tests. Performed on every unit that leaves the factory.
Can shorten or extend the length of the burn-in.  At least
4 hours.

"System design assurance" means testing performed on early units
to make sure the overall design is stable.  We are not engaged
with that department.

Action item: Ray will check to see if we have design assurance
information.

Burn-in: want to stress the machine heavily.  Network, cpu,
memory, vm stress.  One way is to perform kernel compiles.

The burn-in stations will not have peripherals.

Zephaniah: suggests analysis of touch pads pressure sensitivity
and calibration over temperature and unit-to-unit.

Richard: process problems like opens, voids, cracked vias, flux
residue dominate flaky problems.  Dead chips are relatively rare.

Action item: Mitch, Jim, Ray, Mark to decide when Mitch goes to Taiwan.