Gerald A geraldablists at gmail.com
Wed Apr 15 15:52:27 CEST 2009

On Wed, Apr 15, 2009 at 12:03 AM, Werner Almesberger <werner at openmoko.org>wrote:

> Gerald A wrote:
> > >>  - integrate IDBG
> > >
> > >
> > I'm guessing this is a way to unbrick, or a failsafe?
> To unbrick and as a debugging aid (serial console, remote reset,
> and - for those really nasty cases - JTAG). Once you have a
> microcontroller in there, you can also teach it some more tricks
> that are useful for things like testing, such as access to I2C or
> HDQ, or remote activation of the POWER button.

Debugging here is boot debugging, not application debugging, just to
clarify, correct?

> > What I'm wondering is if there would be a way to slap a small Readonly
> piece
> > in there, which can do a reflash. I know I live in utopia here, but it
> might
> > be one way to do it. Comments?
> That's basically what the NOR in GTA02 does. It works quite nicely,
> but it only solves the unbricking, not the debugging, and it sits
> on the memory bus, so it adds to the capacitative load, which can
> limit the maximum memory speed. E.g., GTA02's memory accesses could
> be 33% faster is we didn't have NOR and Glamo on the bus. (Okay,
> without the Glamo, the LCM would have to be refreshed from main
> memory, which basically eats up the speed gain.)

I'm living in utopia, my design has no access compromise. :)
I read something over the last few days about "latching" the boot area, and
I'm sure we
could whip up a linux ISO/CD that allows us to chat with a device on a
standard PC, and
upgrade the boot firmware in a controlled way. So that might work better.
My idea is old-school PC, the way they look for an MBR in a few different
areas (hard disk,
floppy) before launching into the code. We know by now that the boot flash
doesn't change
very often, usually hackers want to change the root and kernel images.

> Question -- is there anything in this form factor which is actually better
> > then the Glamo?
> Yes, no Glamo at all ;-)) The Glamo is essentially an anachronism.
> For that class of devices, graphics acceleration is now implemented
> in the CPU. So any external accelerators will either be very
> specialized or obsolete.
> The 2442 doesn't have a built-in accelerator, but considering that
> memory bandwidth between CPU and main memory is more than an order
> of magnitude better than between CPU and the Glamo framebuffer, even
> unaccelerated graphics would be better in almost all cases than
> using the Glamo.

The Glamo was a headache, no doubt. But trading sharks for boiling oil
always improve your situation. :)

Ok, dumb question: Does this mean we don't need a graphics chip at all?

> > And, if this one is selected, will we be able to do a field upgrade, or
> > would it be another TI GSM firmware upgrade game? (I think not, but just
> > asking).
> Don't TI GSM firmware upgrades work quite well in the field ? And
> yes, upgradeability would be a pretty strong requirement. Once
> bitten, ... :)

I haven't done it yet, but reports are quite positive. That said, it took a
long time
to get that to be a reality, which is why I called it a game. :S

> > Real FPU? No more software floating point? :P
> Perhaps a socket for an external coprocessor, 8087-style ;-)

lol. Musical and you do comedy. :)

> > but if I remember the
> > NDA's were for the documentation provided, but not for the derived works
> > (meaning an OM "employee" could write a document explaining OM's
> > implementation, could write open source drivers and release them, etc).
> That depends on the content of the NDA. Of course, only the kind of
> NDA that lets one write and release Open Source drivers makes sense
> for a project like Openmoko. If there are more freedoms, or if the
> information can be obtained without jumping though hoops, that's of
> course even better.

I agree that more open is better -- but as you've described, we have to be
practical too. A "more open" part with excellent docs is worthless if it
can't scale production to the 10,000 pieces (I'm pulling numbers out of the
air here) needed to be profitable. And an NDA, while definitely annoying,
can still be worked to allow some information to be disclosed and an open
source driver to be written.

And, while 10,000 might seem like a lot to me, companies might think that
amount is slim pickings compared with a commercial run of 1,000,000 or more.
If this project can reach those numbers and keep delivering a consistent
demand, OM's buying leverage will allow you to ask for things those
companies would ignore now.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.openmoko.org/pipermail/gta03/attachments/20090415/80ce8665/attachment.htm 

More information about the Gta03 mailing list