Re: Update on glue code

[Constantine Kousoulos]

Barry deFreese wrote:
> Does this mean you are done with the 2.6 driver code? ;-)
>
> Barry deFreese (aka bddebian)

Hello Barry,

My questions about pthreads came up due to something i happened to
read resently. These questions have nothing to do with driver code
and just fill in some blanks i have about the whole operation of
GNU Hurd.

Regarding the drivers, for the last couple of months i have been
studying the current glue code. My first approach was to see if it
was easy to upgrade the whole glue code to 2.6. While at first
it seemed easy enough, i soon came to realise that something like
that is impossible. The reasons for this are two:
a)There are major architectural differences between the old linux
code and 2.6 that make the transition very difficult.
b)The current glue code is not a distinct layer of code around the
linux sources. Although we tend to think that glue code works
like this:
linux sources  <-->   glue code  <-->  Mach kernel

it is actually implemented like this:
(linux sources mixed with glue code)  <-->  Mach kernel

It is very difficult (if not impossible) to isolate the glue code,
upgrade it and then combine it with the 2.6 drivers.

So, the only thing that someone can do, IMHO, is write a clean
implementation of the glue code that cooperates with 2.6 sources.
In order to do that, you cannot simply start and write "general
purpose" glue code that would theoritically work with all kinds of
drivers. IMHO (again), glue code for each kind of driver should be
written one at a time so that it can be thoroughly debugged and
tested.

The question i've been asking myself lately is what drivers do we
really need developed first? Well, maybe you can help me with that.

Barry, I hope i gave you a clear idea on what's going on. Feel
free to ask me whatever you think i left fuzzy (even if you
haven't "cleaned up" the whole Mach kernel before doing so ;).

Thanks,
Constantine