Marvell microkernel replacement

[David Woodhouse]

On Sun, 2008-01-13 at 02:30 -0500, Albert Cahalan wrote:
> David Woodhouse writes:
> 
> > http://www.csr.com/products/unifirange.htm
> 
> They claim that that is a 1-chip solution. Is it really?

I have no reason to believe otherwise -- why do you ask?

Some people make some fairly preposterous claims in marketing material
but rarely do they make claims which could be so easily disproved. After
all, even the most pointy-haired of managers can usually manage to count
as far as two. :)

> Marvell uses a 2-chip solution.

> If a 2-chip solution is OK, then one could start with a
> 1-chip softmac solution and add any arbitrary processor.
> That CPU could be ARM, MIPS, sh3, sh4, sh5, CRIS,
> ColdFire, Blackfin, 186, PDP-11, IA-64...

Fewer chips is generally better. If we could put the _whole_ thing on
one die -- the kind of thing IBM are really good at doing for their
customers -- then that would be ideal. I don't think we're quite going
to manage _that_ level of integration, but we could certainly do better
than we have in the current XO design.

> In any case:
> 
> At minimum one must get promises in writing, but it's far
> better to have actual published documentation first.
> Don't forget about the errata!

Having seen the kind of NDA and documentation that CSR give to Linux
hackers, I have faith that something entirely acceptable can be worked
out, should we want to go down that path. They very much seem to GetIt™.

-- 
dwmw2