[OLPC-devel] Minutes OLPC BIOS meeting, 2006-08-16

[Mitch Bradley]

OLPC BIOS meeting, 2006-08-16

Present: Richard, Jim, Jordan, Ray, Marcello, Mitch (scribe)

Jordan - definitely having audio problem "Couldn't initialize codec".
Doesn't think it's the interrupt problem, but not sure what it is.
It's as if the codec is not there.  Perhaps LinuxBIOS is not turning
on something magic.

Action item for Jordan - post the bug report to the devel list.

Marcello says that harmless codec read failures have occurred in the past.

Need to give Jayah a head's up about the situation.

State of fbdev driver:  The NOVGA patch irritates Qemu.  The fbdev 
driver is no
problem.  Paranoia about VGA.  Decision: don't worry about Qemu; 
optimize for
the OLPC hardware.

Regarding the Insyde BIOS expiration (on Aug 26): we would like to be able
to switch over to LinuxBIOS before then, without interfering with anybody's
development.  There is a workaround - use the BIOS setup screen to turn back
the clock, and use NTP to set the clock under Linux.  We need to get people
switched over to LinuxBIOS to flush out bugs.

Xorg.conf - option nocompression need to be turned on.  But the compression
forces a power of 2 pitch, which causes each line to take up 8K, which 
wastes
a lot of memory on OLPC (about 4M).  But the compression really helps from a
bandwidth standpoint.  So most other platforms would prefer compression.

Jim's take: for now, compress to save bandwidth, burn the space.

Which display mode?  Patch has been sent to David.  1200x900 mode is 
hijacked
to be either a VGA (85Hz) version or a DCON (50Hz) version.

Action item: Need to recalculate the VGA frequency - 85 is too high.

Mary Lou is scheduled to arrive with a DCON tomorrow.

PS/2 works on LinuxBIOS but not on Insyde.

olpcflash - takes 11 minutes to FLASH SPI.  Ray says it should take 1 
minute.
The part data sheet says that it takes 1.4 ms to program 256 bytes.
That works out to 6 seconds for 1 MByte.  Clocking in the Page Program 
instruction
with 256 bytes of payload takes 32 + (256*8) = 2079 clocks.  At 1 MHz 
SPI clock
rate, that would add about 8 seconds to the time (neglecting the shorter
miscellaneous setup writes).  At 25 MHz, the extra time would be 
insignificant.

We need to check for bogus images before committing the FLASH write.
Discussion of techniques.  Some combination of looking for a signature
in the ID block and checksumming.

Ray wants an option to program the EC area too.  Richard will call it the
--brick option.

AIs for Richard: image verification, --brick option, speed up programming.

Pulling KBC out of reset after programming SPI causes system reset.  
Rays says
that the DOS SPI programming utility displays a message telling you to 
reboot
the system.  Ray has a version of Insyde BIOS that works with PS/2.

Action item for Ray: feed back more information about keyboard reset.

We have had a number of requests from Brazil, so we could use a Portugese
translation of the instructions.

Marcello: ship buildrom binary in Fedora.

Mitch: suggested a dd-able installation.

Marcello: something like that exists

Mitch: until it's documented on the wiki, it doesn't count

Jim: will get someone going on an easy installation procedure directly to
a 256MB (or so) USB key.