[OLPC-devel] Minutes: BIOS teleconference Wednesday Evening Boston Time.

[Mitch Bradley]

Minutes for OLPC BIOS meeting, 2006-08-02, taken by Mitch Bradley

This meeting was split out from a meeting that occurred on 2006-08-02.


  Agenda
  ======
  Please send me any items to add to the list.  These occur to me as
  topics we should start working through, get owners for and get closed.
  It is a long list; we'll have to be disciplined to get through it.
 
  LinuxBIOS status
  Which kernel code base to use: kernel.org vs. Fedora? (kernel.org, 
probably)
  How to merge OLPC specific drivers into this code base, until they are 
upstream.
  How to archive "releases" of the bios/Linux as bootloader?
  How to archive corresponding TinyLinux patches?
  Config file to use.
  Flash update utility status/needs
  Compression status / TCP/IP stack status / wireless install?
  Is mem= needed? Jordan thinks yes.
  Can we autodetect memory size/flash size?
  User interface.
  Requirements to make production as fast/cheap/easy as possible.
  EC code development status.
  EC code update utility.
  Serial number handling.
  Verification process before commitment to BTest.

Present:

Mitch Bradley
Jim Gettys
Ron Minnich (LinuxBIOS - mediates commits)
Richard Smith (LinuxBIOS - helps Ron)
Ray (Quanta)


Jim introduced Mitch Bradley, who is new to the project.  Mitch has been
doing firmware and drivers for many years, much of it in the context of
OpenBoot on Suns and Open Firmware on other systems.  Some of the 
information
below is probably well known to everyone; it is in response to Mitch's
coming-up-to-speed questions.

Jim: Marcello had a accident and requires leg surgery.  He may be out of
commission for several weeks if power pain killers are involved.

Expect to contract with Jonathan Corbet (Linux device driver expert).


** LinuxBIOS status:

Data compression: Need to get back a couple of K.  Can save 20K by 
compressing
a different way.

We really want to do LZMA compression, which would be even better.

Need more study about how to do S3 state without ACPI etc.  Strawman: 
don't use ACPI.

Jim: Two ACPI functions need to be separately considered:
  a) User space access to e.g. battery status
  b) how to suspend/resume.
May want to use ACPI for a but not b.

Conjecture: The reason that RedHat kernels don't work with LinuxBIOS
could be because of dependencies on SMM mode.

Steve Goodrich is no longer at AMD and is looking for a job.  Nicholas 
is meeting
with Hector Ruiz (who is busy with ATI acquisition), trying get us access to
someone else.

We can't compile the VSA code because you need a 16-bit C compiler with
support for specific switches.

AMD is putting the Geode line into maintenance mode.  No roadmap past LX.

Close to having install onto FLASH from a USB key.  Need to:
a) Install LinuxBIOS into FLASH.
b) Install Linux onto NAND FLASH.


** Kernel code base:

Suggestion: Let's go with kernel.org and use truncated config.

Can't kexec Fedora kernels.

Jordan Krause is an AMD guy who survived layoffs - fbdev and X and 
general maintenance.
cosmic penguin

buildROM pulls 6.18rc2 from Jordan and it is working well.

Need to update fbdev driver to support DCON.

News: OLPC dual-mode display is working.  No pictures yet.  Mary Lou thinks
she can double the reflectance.


** How to archive "releases" of the bios/Linux as bootloader?

We need to generate a bootrom tarball that has all the bits corresponding
to binary image in order to obey the GPL.

Something like:  make mrproper; tar ...


** How to archive corresponding TinyLinux patches?

Need to check patch set versus .18 to see what has gone upstream.


** Config file to use for LinuxBIOS

Need LZMA and NAND driver ... and then CaFe NAND driver.

The CaFe chip supports NAND FLASH, Omnivision camera, SDIO (SD HCI from 
Marvell,
driver already exists).  Jonathan Corbet thinks that the camera chip is
similar to earlier chips, but needs new host controller.  Existing 
Omnivision
drivers will be close.


** Compression status / TCP/IP stack status / wireless install?

Kernel mods for LZMA are available.  LZMA seems to do better than gzip
for executables.  Saves ~190K.  May buy us enough for TCP stack.
Fallback is to put the TCP stuff as a separate module in NAND FLASH.

We need to draw the line soon so we can freeze for the beta build.

AI (Jim and Ray): Write a boot requirements document.  Ray will write draft,
Jim will review.

Decision: defer compression changes until after beta, to reduce risk.
Focus on functionality first.  Put TCP module in NAND FLASH for now.

First step: make it possible to program SPI ROM.


-- At this point the meeting adjourned, tabling the additional agenda items
until later.  Ron had to leave early, and submitted comments on those
additional agenda items via email, see below:



** Is mem= needed? Jordan thinks yes.

Ron: I think it's only needed due to a bug in the driver or the way the 
kernel comes up. I have no objection to it.


** Can we autodetect memory size/flash size?

Ron: yes in both cases, at what point in bootup do you wish to do this?


** User interface.

Ron: Not sure which UI you mean  :-)


** Requirements to make production as fast/cheap/easy as possible.

Ron: we have to pick the usb->enet dongle so we can start testing it.