Re: userspace drivers

[Niels Möller]

Marcus Brinkmann <Marcus.Brinkmann@ruhr-uni-bochum.de> writes:

> Peter has already written device drivers for embedded Linux, and actually
> has a lot of experience on how to write device drivers portably (he is
> actually giving a talk at FOSDEM about it, he announced this on this list).

That's fine. Hurd requirements might be a little different, though.

> A proper driver framework is critical for extensibility and performance. 
> The write a few drivers first and then fix it later is how Linus did it,
> and look at the result. They are still changing and changing the framework,
> and always have to fix all those drivers to follow these changes.

Well, large numbers of user space drivers will, I hope, not happen
until L4. I'm only suggesting that writing a few (say, at *most* half
a dozen) drivers for GNU Mach will provide experience that will be
valuable when moving to L4 for real.

Even when you have designed and implemented a driver framework, the
first thing you would do is to write a few drivers. And when you do
that, you'll most likely discover one or two bugs in the interface,
and then modify the framework and drivers as necessary to correct it.
The interface will need some time and a few drivers of different types
in order to mature.

I tried to review the discussion on l4-hurd (starting with
<874ra4owhx.fsf@monom.org>, Daniel Wagner, Tue, 26 Nov 2002). Most of
that is about the central book-keeping of available hardware
resources. The impact of that design onto an individual driver should
be pretty small. In particular for a parallell port driver, which I
imagine (without much knowledge of PC hardware) will not need to know
much about the various i/o buses.

/Niels