[OLPC-devel] System software meeting minutes, 2006-09-05

[Mitch Bradley]

System Software telecon minutes, 2006-09-05

Jim Gettys, Jordan Crouse, Ray, Victor Chau - Quanta project mgr in charge
of OLPC R&D, Jon Corbett, David Woodhouse, Chris Blizzard, Mitch Bradley,
Zephaniah Hull, Tom Sylla, Marcelo, Richard Smith

CaFe:

CaFe test hardware - one copy to Pierre Ossman (sd maintainer), in 
addition to
the copies to dave and mitch

Pierre to test Marvell's reimplementation to the public spec.

Jon to hook up with Dave on Wednesday night.  Mitch leaves early 
Wednesday morning so hookup is dubious.

NAND is largely working, mounting and working with programmed I/O.  Dave
doesn't have DMA going yet.

David thinks we don't need any special initialization for CaFe in LinuxBIOS.

Tom Sylla - performance optimization opportunities via region configuration
of BAR-mapped regions.

USB Ethernet performance hit:

Lots of ideas but nobody actually knows anything for sure.

Marcelo: Perhaps it is a particularly stupid device.

Jim: it affects our wireless.

Possibly an interrupt storm caused by status interrupt rearming itself.

Marcelo suggests James Morris as a good person to look at it.

AI: Chris to contact James Morris.

DCON driver:

Jordan wrote some code to enable DCON driver via SMBus.

DCONs got busted in shipment.  0 for 3 so far.

New shipments going out tomorrow.  They need to be marked as engineering
samples, otherwise we get

AI: Jim G. to check DCON shipments - mark as "Engineering Sample, No
Commercial Value" - and status of Dave's DCON.

Richard: No reports of bricked machines from LinuxBIOS upgrades.

Discussion of procedure for testing Marcelo's bad boards.  Marcelo
suggests waiting for JTAG board.  Flashing Insyde would be a good test.
Zephaniah suggests booting with LB on a PLCC.  But Marcelo has tried
with a known good LB PLCC FLASH.

Possibility of boards degrading over time.  How can we do failure analysis?

Mitch can do software-based debugging using ROM emulator.

FS2 is the answer.

AI: Mitch - Tell Mark Foster to send one FS2 cable to Cambridge, one to 
Marcello
AI: Jim - make sure we get enough cables for JTAG on Btest boards.

Jordan's failure case - all known info is in the trac report.  He will get
a special cable made tomorrow to handle the bent-over pins.

Zephaniah - cables have arrived, waiting for Alps samples.  Jim is 
bugging Alps. They were supposed to ship end of last month.  Issue with 
pressure sensing - only 6 values.  The range is not great - a small 
pressure maxes it out.

AI: Jim to continue pinging Alps.

Schedules:

Some Btest boards with FPGA for CaFE late Sept.  DCON ASICs for verification
Oct. 24.  Btest build of systems - 1000 units mid/late Nov (CaFe FPGA),
4000 machines with CaFe ASICs in January.  C build a couple of months 
later -
identical to production.  Full production in April.  Batteries will be
available for September machines.  Btest boards will be ready for Linux
driver work around battery life.

Victor responsible for making sure we have all battery stuff.

Quanta needs power management.  Conflict with Mitch's proposed trip to
Taiwan.

Mitch to work on some sort of diagnostic for power management.

David: Linux has a full-system suspend capability, but no fine-grained
power management.

1) Enumerate the power domain
2) Implement shutdown for individual power domains
3) Implement the full system elaboration

AI: Mitch create a strawman for a power management diag framework and
start a list of power domains.  - buses, clocks, devices, yada yada

PS2 - Can't cut off PS2 power in some modes, because of DCON wakeup via
keyboard and touchpad.  You can turn off the PS2 when the lid is closed.