[geeks] A Real OS? (was: Re: my capitalization.. etc.)

[Jonathan C. Patschke]

On Sun, 19 May 2002, Eric Dittman wrote:

> Since when is nVidia some "Joe Bob's Backwoods" video card?
> And do you really think that having more vendors supply drivers
> for their hardware would be bad for *BSD?

Kernel-mode drivers?  Damn stright it's a bad idea.  Take a look at
Windows NT 4.0 and later.  Funny how I -never- got a bluescreen under NT
until they let companies other than Microsoft play around under the hood.
I always ran hardware on the HCL, and, with 3.51 and before, it worked
perfectly.  With 4.0 and later, things just went to hell.

If the users want drivers, let them confince manufacturers work with
XFree86 to provide servers that support XGL and whatever else, and then
let the OS present a seomwhat generic interface to AGP/PCI/VESA for such
servers to use.  Supporting every possible video, audio, etc. card in the
kernel of the OS is a really stupid idea.

> Of course in the nVidia case we are not talking kernel about
> kernel code but rather XFree86.

The last nVidia driver I used for Linux required the insertion of a kernel
module.  If the nVidia driver -isn't- a kernel module, there's no reason
it cannot work on *BSD, provided that the *BSD kernel exposes the proper
buses.  That's a vendor issue, not an OS issue, and thanks to all the
vendors today being tight-lipped about how their hardware works, it's
Really Difficult for anyone to fix that.

Which basically amounts this situation:
  1) OS needs hardware support to gain larger numbers of users.
  2) Hardware companies only support OSes with large numbers of users.

Catch-22.

> I can't see where limited hardware support will help wider
> acceptance of *BSD.  Look at what limited hardware support
> has done to Solaris x86.

I always thought it was the fact that it ran like ass.  The first install
I did on an dual P90 took forty -hours-.

> I really can't believe you want limited hardware support
> in *BSD.

Not limited hardware support in general.  He's (I think) advocating
limited hardware support in the -kernel-.  At least, that's what I'm
advocating.

> If the drivers are well written and tested before inclusion
> there wouldn't be a problem.

To properly test kernel-level code, you need to test every possible
permutation, which is Not Easy.  Otherwise, you have nVidia's driver
stomping all over the DMA access that Creative Labs' driver needs because
NEITHER of those companies can write a decent driver to save their lives.
I can consistently crash my SB Live! driver for Win2k, and my nVidia
driver, both by not doing anything out of the ordinary.

That's the primary point here, and it's a point that was swung completely
in the other direction before the introduction of Windows 3.0.  Back then
the hardware developers didn't write -any- drivers--the just provided
specifications so that the people that knew the operating systems best
could provide hardware support.  Now, the OS folks have to open up their
APIs to the hardawre guys so that they can write half-assed interface code
that bombs all the time, or adds "neat" little features here and there to
bloat things up.

--Jonathan