From: Emmanuel Marty <core@suntech.fr>
  To  : ggi-develop@eskimo.com
  Date: Fri, 12 Feb 1999 00:17:33 +0100

Re: STB and S3 tech specs

Marcus Sundberg wrote:

> Jon M. Taylor wrote:
> >         Because releasing their specs would allow their competitors to
> > reverse-engineer their hardware.  If you know the register layout, you can
> > hook up a logic analyzer to the chip(s) and discover exactly what the
> > internals of the chip are.  EXACTLY, down to the individual logic gates!
> > And since that layout is a trade secret and not patented, if your
> > competitors figure it out they can clone it 100%, not pay you anything,
> > and get rich.  This _has_ happened, exactly as I described above, many
> > times in the past.
> 
> Hmm, I would have thought that disassembling and understanding a
> binary driver would not take any time worth mentioning compared to
> reverse-engineering a chip?

That's my opinion as well. I am used to debug at cpu instruction
level anyway (and admittedly I even like it :) and I am in Europe, so
this would be a perfectly possible option. I believe Andy wouldn't
have problems with that either, nor would you Marcus.. And a lot
more people could.

That, and if it's a binary KGI driver, it would take about half
an hour to modify suidkgi to trap and print all accesses to i/o ports
and give the driver not the mmio areas it asked for, but fake
unmapped ones that also get faulted and trapped and then the values
are read or written in the actual memory. I'm sure most of the
KGI driver writers could see patterns and figure out what it does
when testing the various display mode settings, polygon fillings,
etc. Shall we propose a challenge to creative? ;) "if we understand
what it does, release the real source"? :-)

Cyrix made that reasoning and let the source out. I'm not sure
it made them sell more mediaGX chips because of that, but I know that
we use it in all embedded products we are developing at work now;
it's been over a year since I was allowed to release the driver
sources, and I still have to hear from a competitor clone. Sure this
is a niche market, but this is a LARGE niche market. If a competitor
ever clones the mediaGxm display functions in the next 6 months, cyrix
will be releasing a new chip by that time for sure.

>From late messages on the linux-kernel ML, linus is both opposed to
supporting binary-only drivers and allowing their writers to make
wrappers (like KGI) even if themselves they are GPLed. Not sure how
the drivers will become mainstream in this situation :) I can
somehow foresee a "linux with real hardware support but giving up
on what took it there in the first place" fork in the near future..

This said, I still prefer binary drivers than none at all. And considering
that Creative hired Jon of all people, I am very confident that we
will have excellent quality drivers for our favorite system soon. I have
no religious opposition to closed-source development either; I just
think it's not going to be evident to integrate it into a system that
was built openly from the ground up :) Best wishes to Jon to pull that
stunt and release quality drivers for hardware some of us are panting
at anyway :)

--
Emmanuel